1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _NET_NEIGHBOUR_H
3#define _NET_NEIGHBOUR_H
4
5#include <linux/neighbour.h>
6
7/*
8 * Generic neighbour manipulation
9 *
10 * Authors:
11 * Pedro Roque <roque@di.fc.ul.pt>
12 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
13 *
14 * Changes:
15 *
16 * Harald Welte: <laforge@gnumonks.org>
17 * - Add neighbour cache statistics like rtstat
18 */
19
20#include <linux/atomic.h>
21#include <linux/refcount.h>
22#include <linux/netdevice.h>
23#include <linux/skbuff.h>
24#include <linux/rcupdate.h>
25#include <linux/seq_file.h>
26#include <linux/bitmap.h>
27
28#include <linux/err.h>
29#include <linux/sysctl.h>
30#include <linux/workqueue.h>
31#include <net/rtnetlink.h>
32#include <net/neighbour_tables.h>
33
34/*
35 * NUD stands for "neighbor unreachability detection"
36 */
37
38#define NUD_IN_TIMER (NUD_INCOMPLETE|NUD_REACHABLE|NUD_DELAY|NUD_PROBE)
39#define NUD_VALID (NUD_PERMANENT|NUD_NOARP|NUD_REACHABLE|NUD_PROBE|NUD_STALE|NUD_DELAY)
40#define NUD_CONNECTED (NUD_PERMANENT|NUD_NOARP|NUD_REACHABLE)
41
42struct neighbour;
43
44enum {
45 NEIGH_VAR_MCAST_PROBES,
46 NEIGH_VAR_UCAST_PROBES,
47 NEIGH_VAR_APP_PROBES,
48 NEIGH_VAR_MCAST_REPROBES,
49 NEIGH_VAR_RETRANS_TIME,
50 NEIGH_VAR_BASE_REACHABLE_TIME,
51 NEIGH_VAR_DELAY_PROBE_TIME,
52 NEIGH_VAR_INTERVAL_PROBE_TIME_MS,
53 NEIGH_VAR_GC_STALETIME,
54 NEIGH_VAR_QUEUE_LEN_BYTES,
55 NEIGH_VAR_PROXY_QLEN,
56 NEIGH_VAR_ANYCAST_DELAY,
57 NEIGH_VAR_PROXY_DELAY,
58 NEIGH_VAR_LOCKTIME,
59#define NEIGH_VAR_DATA_MAX (NEIGH_VAR_LOCKTIME + 1)
60 /* Following are used as a second way to access one of the above */
61 NEIGH_VAR_QUEUE_LEN, /* same data as NEIGH_VAR_QUEUE_LEN_BYTES */
62 NEIGH_VAR_RETRANS_TIME_MS, /* same data as NEIGH_VAR_RETRANS_TIME */
63 NEIGH_VAR_BASE_REACHABLE_TIME_MS, /* same data as NEIGH_VAR_BASE_REACHABLE_TIME */
64 /* Following are used by "default" only */
65 NEIGH_VAR_GC_INTERVAL,
66 NEIGH_VAR_GC_THRESH1,
67 NEIGH_VAR_GC_THRESH2,
68 NEIGH_VAR_GC_THRESH3,
69 NEIGH_VAR_MAX
70};
71
72struct neigh_parms {
73 possible_net_t net;
74 struct net_device *dev;
75 netdevice_tracker dev_tracker;
76 struct list_head list;
77 int (*neigh_setup)(struct neighbour *);
78 struct neigh_table *tbl;
79
80 void *sysctl_table;
81
82 int dead;
83 refcount_t refcnt;
84 struct rcu_head rcu_head;
85
86 int reachable_time;
87 u32 qlen;
88 int data[NEIGH_VAR_DATA_MAX];
89 DECLARE_BITMAP(data_state, NEIGH_VAR_DATA_MAX);
90};
91
92static inline void neigh_var_set(struct neigh_parms *p, int index, int val)
93{
94 set_bit(nr: index, addr: p->data_state);
95 p->data[index] = val;
96}
97
98#define NEIGH_VAR(p, attr) ((p)->data[NEIGH_VAR_ ## attr])
99
100/* In ndo_neigh_setup, NEIGH_VAR_INIT should be used.
101 * In other cases, NEIGH_VAR_SET should be used.
102 */
103#define NEIGH_VAR_INIT(p, attr, val) (NEIGH_VAR(p, attr) = val)
104#define NEIGH_VAR_SET(p, attr, val) neigh_var_set(p, NEIGH_VAR_ ## attr, val)
105
106static inline void neigh_parms_data_state_setall(struct neigh_parms *p)
107{
108 bitmap_fill(dst: p->data_state, NEIGH_VAR_DATA_MAX);
109}
110
111static inline void neigh_parms_data_state_cleanall(struct neigh_parms *p)
112{
113 bitmap_zero(dst: p->data_state, NEIGH_VAR_DATA_MAX);
114}
115
116struct neigh_statistics {
117 unsigned long allocs; /* number of allocated neighs */
118 unsigned long destroys; /* number of destroyed neighs */
119 unsigned long hash_grows; /* number of hash resizes */
120
121 unsigned long res_failed; /* number of failed resolutions */
122
123 unsigned long lookups; /* number of lookups */
124 unsigned long hits; /* number of hits (among lookups) */
125
126 unsigned long rcv_probes_mcast; /* number of received mcast ipv6 */
127 unsigned long rcv_probes_ucast; /* number of received ucast ipv6 */
128
129 unsigned long periodic_gc_runs; /* number of periodic GC runs */
130 unsigned long forced_gc_runs; /* number of forced GC runs */
131
132 unsigned long unres_discards; /* number of unresolved drops */
133 unsigned long table_fulls; /* times even gc couldn't help */
134};
135
136#define NEIGH_CACHE_STAT_INC(tbl, field) this_cpu_inc((tbl)->stats->field)
137
138struct neighbour {
139 struct hlist_node hash;
140 struct hlist_node dev_list;
141 struct neigh_table *tbl;
142 struct neigh_parms *parms;
143 unsigned long confirmed;
144 unsigned long updated;
145 rwlock_t lock;
146 refcount_t refcnt;
147 unsigned int arp_queue_len_bytes;
148 struct sk_buff_head arp_queue;
149 struct timer_list timer;
150 unsigned long used;
151 atomic_t probes;
152 u8 nud_state;
153 u8 type;
154 u8 dead;
155 u8 protocol;
156 u32 flags;
157 seqlock_t ha_lock;
158 unsigned char ha[ALIGN(MAX_ADDR_LEN, sizeof(unsigned long))] __aligned(8);
159 struct hh_cache hh;
160 int (*output)(struct neighbour *, struct sk_buff *);
161 const struct neigh_ops *ops;
162 struct list_head gc_list;
163 struct list_head managed_list;
164 struct rcu_head rcu;
165 struct net_device *dev;
166 netdevice_tracker dev_tracker;
167 u8 primary_key[];
168} __randomize_layout;
169
170struct neigh_ops {
171 int family;
172 void (*solicit)(struct neighbour *, struct sk_buff *);
173 void (*error_report)(struct neighbour *, struct sk_buff *);
174 int (*output)(struct neighbour *, struct sk_buff *);
175 int (*connected_output)(struct neighbour *, struct sk_buff *);
176};
177
178struct pneigh_entry {
179 struct pneigh_entry __rcu *next;
180 possible_net_t net;
181 struct net_device *dev;
182 netdevice_tracker dev_tracker;
183 union {
184 struct list_head free_node;
185 struct rcu_head rcu;
186 };
187 u32 flags;
188 u8 protocol;
189 bool permanent;
190 u32 key[];
191};
192
193/*
194 * neighbour table manipulation
195 */
196
197#define NEIGH_NUM_HASH_RND 4
198
199struct neigh_hash_table {
200 struct hlist_head *hash_heads;
201 unsigned int hash_shift;
202 __u32 hash_rnd[NEIGH_NUM_HASH_RND];
203 struct rcu_head rcu;
204};
205
206
207struct neigh_table {
208 int family;
209 unsigned int entry_size;
210 unsigned int key_len;
211 __be16 protocol;
212 __u32 (*hash)(const void *pkey,
213 const struct net_device *dev,
214 __u32 *hash_rnd);
215 bool (*key_eq)(const struct neighbour *, const void *pkey);
216 int (*constructor)(struct neighbour *);
217 int (*pconstructor)(struct pneigh_entry *);
218 void (*pdestructor)(struct pneigh_entry *);
219 void (*proxy_redo)(struct sk_buff *skb);
220 int (*is_multicast)(const void *pkey);
221 bool (*allow_add)(const struct net_device *dev,
222 struct netlink_ext_ack *extack);
223 char *id;
224 struct neigh_parms parms;
225 struct list_head parms_list;
226 int gc_interval;
227 int gc_thresh1;
228 int gc_thresh2;
229 int gc_thresh3;
230 unsigned long last_flush;
231 struct delayed_work gc_work;
232 struct delayed_work managed_work;
233 struct timer_list proxy_timer;
234 struct sk_buff_head proxy_queue;
235 atomic_t entries;
236 atomic_t gc_entries;
237 struct list_head gc_list;
238 struct list_head managed_list;
239 rwlock_t lock;
240 unsigned long last_rand;
241 struct neigh_statistics __percpu *stats;
242 struct neigh_hash_table __rcu *nht;
243 struct mutex phash_lock;
244 struct pneigh_entry __rcu **phash_buckets;
245};
246
247static inline int neigh_parms_family(struct neigh_parms *p)
248{
249 return p->tbl->family;
250}
251
252#define NEIGH_PRIV_ALIGN sizeof(long long)
253#define NEIGH_ENTRY_SIZE(size) ALIGN((size), NEIGH_PRIV_ALIGN)
254
255static inline void *neighbour_priv(const struct neighbour *n)
256{
257 return (char *)n + n->tbl->entry_size;
258}
259
260/* flags for neigh_update() */
261#define NEIGH_UPDATE_F_OVERRIDE BIT(0)
262#define NEIGH_UPDATE_F_WEAK_OVERRIDE BIT(1)
263#define NEIGH_UPDATE_F_OVERRIDE_ISROUTER BIT(2)
264#define NEIGH_UPDATE_F_USE BIT(3)
265#define NEIGH_UPDATE_F_MANAGED BIT(4)
266#define NEIGH_UPDATE_F_EXT_LEARNED BIT(5)
267#define NEIGH_UPDATE_F_ISROUTER BIT(6)
268#define NEIGH_UPDATE_F_ADMIN BIT(7)
269#define NEIGH_UPDATE_F_EXT_VALIDATED BIT(8)
270
271/* In-kernel representation for NDA_FLAGS_EXT flags: */
272#define NTF_OLD_MASK 0xff
273#define NTF_EXT_SHIFT 8
274#define NTF_EXT_MASK (NTF_EXT_MANAGED | NTF_EXT_EXT_VALIDATED)
275
276#define NTF_MANAGED (NTF_EXT_MANAGED << NTF_EXT_SHIFT)
277#define NTF_EXT_VALIDATED (NTF_EXT_EXT_VALIDATED << NTF_EXT_SHIFT)
278
279extern const struct nla_policy nda_policy[];
280
281#define neigh_for_each_in_bucket(pos, head) hlist_for_each_entry(pos, head, hash)
282#define neigh_for_each_in_bucket_rcu(pos, head) \
283 hlist_for_each_entry_rcu(pos, head, hash)
284#define neigh_for_each_in_bucket_safe(pos, tmp, head) \
285 hlist_for_each_entry_safe(pos, tmp, head, hash)
286
287static inline bool neigh_key_eq32(const struct neighbour *n, const void *pkey)
288{
289 return *(const u32 *)n->primary_key == *(const u32 *)pkey;
290}
291
292static inline bool neigh_key_eq128(const struct neighbour *n, const void *pkey)
293{
294 const u32 *n32 = (const u32 *)n->primary_key;
295 const u32 *p32 = pkey;
296
297 return ((n32[0] ^ p32[0]) | (n32[1] ^ p32[1]) |
298 (n32[2] ^ p32[2]) | (n32[3] ^ p32[3])) == 0;
299}
300
301static inline struct neighbour *___neigh_lookup_noref(
302 struct neigh_table *tbl,
303 bool (*key_eq)(const struct neighbour *n, const void *pkey),
304 __u32 (*hash)(const void *pkey,
305 const struct net_device *dev,
306 __u32 *hash_rnd),
307 const void *pkey,
308 struct net_device *dev)
309{
310 struct neigh_hash_table *nht = rcu_dereference(tbl->nht);
311 struct neighbour *n;
312 u32 hash_val;
313
314 hash_val = hash(pkey, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
315 neigh_for_each_in_bucket_rcu(n, &nht->hash_heads[hash_val])
316 if (n->dev == dev && key_eq(n, pkey))
317 return n;
318
319 return NULL;
320}
321
322static inline struct neighbour *__neigh_lookup_noref(struct neigh_table *tbl,
323 const void *pkey,
324 struct net_device *dev)
325{
326 return ___neigh_lookup_noref(tbl, key_eq: tbl->key_eq, hash: tbl->hash, pkey, dev);
327}
328
329static inline void neigh_confirm(struct neighbour *n)
330{
331 if (n) {
332 unsigned long now = jiffies;
333
334 /* avoid dirtying neighbour */
335 if (READ_ONCE(n->confirmed) != now)
336 WRITE_ONCE(n->confirmed, now);
337 }
338}
339
340void neigh_table_init(int index, struct neigh_table *tbl);
341int neigh_table_clear(int index, struct neigh_table *tbl);
342struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
343 struct net_device *dev);
344struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
345 struct net_device *dev, bool want_ref);
346static inline struct neighbour *neigh_create(struct neigh_table *tbl,
347 const void *pkey,
348 struct net_device *dev)
349{
350 return __neigh_create(tbl, pkey, dev, want_ref: true);
351}
352void neigh_destroy(struct neighbour *neigh);
353int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb,
354 const bool immediate_ok);
355int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new, u32 flags,
356 u32 nlmsg_pid);
357void __neigh_set_probe_once(struct neighbour *neigh);
358bool neigh_remove_one(struct neighbour *ndel);
359void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev);
360int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev);
361int neigh_carrier_down(struct neigh_table *tbl, struct net_device *dev);
362int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb);
363int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb);
364int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb);
365struct neighbour *neigh_event_ns(struct neigh_table *tbl,
366 u8 *lladdr, void *saddr,
367 struct net_device *dev);
368
369struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
370 struct neigh_table *tbl);
371void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms);
372
373static inline
374struct net *neigh_parms_net(const struct neigh_parms *parms)
375{
376 return read_pnet(pnet: &parms->net);
377}
378
379unsigned long neigh_rand_reach_time(unsigned long base);
380
381void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
382 struct sk_buff *skb);
383struct pneigh_entry *pneigh_lookup(struct neigh_table *tbl, struct net *net,
384 const void *key, struct net_device *dev);
385int pneigh_create(struct neigh_table *tbl, struct net *net, const void *key,
386 struct net_device *dev, u32 flags, u8 protocol,
387 bool permanent);
388int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *key,
389 struct net_device *dev);
390
391static inline struct net *pneigh_net(const struct pneigh_entry *pneigh)
392{
393 return read_pnet(pnet: &pneigh->net);
394}
395
396void neigh_app_ns(struct neighbour *n);
397void neigh_for_each(struct neigh_table *tbl,
398 void (*cb)(struct neighbour *, void *), void *cookie);
399void __neigh_for_each_release(struct neigh_table *tbl,
400 int (*cb)(struct neighbour *));
401int neigh_xmit(int fam, struct net_device *, const void *, struct sk_buff *);
402
403struct neigh_seq_state {
404 struct seq_net_private p;
405 struct neigh_table *tbl;
406 struct neigh_hash_table *nht;
407 void *(*neigh_sub_iter)(struct neigh_seq_state *state,
408 struct neighbour *n, loff_t *pos);
409 unsigned int bucket;
410 unsigned int flags;
411#define NEIGH_SEQ_NEIGH_ONLY 0x00000001
412#define NEIGH_SEQ_IS_PNEIGH 0x00000002
413#define NEIGH_SEQ_SKIP_NOARP 0x00000004
414};
415void *neigh_seq_start(struct seq_file *, loff_t *, struct neigh_table *,
416 unsigned int);
417void *neigh_seq_next(struct seq_file *, void *, loff_t *);
418void neigh_seq_stop(struct seq_file *, void *);
419
420int neigh_proc_dointvec(const struct ctl_table *ctl, int write,
421 void *buffer, size_t *lenp, loff_t *ppos);
422int neigh_proc_dointvec_jiffies(const struct ctl_table *ctl, int write,
423 void *buffer,
424 size_t *lenp, loff_t *ppos);
425int neigh_proc_dointvec_ms_jiffies(const struct ctl_table *ctl, int write,
426 void *buffer, size_t *lenp, loff_t *ppos);
427
428int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
429 proc_handler *proc_handler);
430void neigh_sysctl_unregister(struct neigh_parms *p);
431
432static inline void __neigh_parms_put(struct neigh_parms *parms)
433{
434 refcount_dec(r: &parms->refcnt);
435}
436
437static inline struct neigh_parms *neigh_parms_clone(struct neigh_parms *parms)
438{
439 refcount_inc(r: &parms->refcnt);
440 return parms;
441}
442
443/*
444 * Neighbour references
445 */
446
447static inline void neigh_release(struct neighbour *neigh)
448{
449 if (refcount_dec_and_test(r: &neigh->refcnt))
450 neigh_destroy(neigh);
451}
452
453static inline struct neighbour * neigh_clone(struct neighbour *neigh)
454{
455 if (neigh)
456 refcount_inc(r: &neigh->refcnt);
457 return neigh;
458}
459
460#define neigh_hold(n) refcount_inc(&(n)->refcnt)
461
462static __always_inline int neigh_event_send_probe(struct neighbour *neigh,
463 struct sk_buff *skb,
464 const bool immediate_ok)
465{
466 unsigned long now = jiffies;
467
468 if (READ_ONCE(neigh->used) != now)
469 WRITE_ONCE(neigh->used, now);
470 if (!(READ_ONCE(neigh->nud_state) & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE)))
471 return __neigh_event_send(neigh, skb, immediate_ok);
472 return 0;
473}
474
475static inline int neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
476{
477 return neigh_event_send_probe(neigh, skb, immediate_ok: true);
478}
479
480#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
481static inline int neigh_hh_bridge(struct hh_cache *hh, struct sk_buff *skb)
482{
483 unsigned int seq, hh_alen;
484
485 do {
486 seq = read_seqbegin(&hh->hh_lock);
487 hh_alen = HH_DATA_ALIGN(ETH_HLEN);
488 memcpy(skb->data - hh_alen, hh->hh_data, ETH_ALEN + hh_alen - ETH_HLEN);
489 } while (read_seqretry(&hh->hh_lock, seq));
490 return 0;
491}
492#endif
493
494static inline int neigh_hh_output(const struct hh_cache *hh, struct sk_buff *skb)
495{
496 unsigned int hh_alen = 0;
497 unsigned int seq;
498 unsigned int hh_len;
499
500 do {
501 seq = read_seqbegin(sl: &hh->hh_lock);
502 hh_len = READ_ONCE(hh->hh_len);
503 if (likely(hh_len <= HH_DATA_MOD)) {
504 hh_alen = HH_DATA_MOD;
505
506 /* skb_push() would proceed silently if we have room for
507 * the unaligned size but not for the aligned size:
508 * check headroom explicitly.
509 */
510 if (likely(skb_headroom(skb) >= HH_DATA_MOD)) {
511 /* this is inlined by gcc */
512 memcpy(to: skb->data - HH_DATA_MOD, from: hh->hh_data,
513 HH_DATA_MOD);
514 }
515 } else {
516 hh_alen = HH_DATA_ALIGN(hh_len);
517
518 if (likely(skb_headroom(skb) >= hh_alen)) {
519 memcpy(to: skb->data - hh_alen, from: hh->hh_data,
520 len: hh_alen);
521 }
522 }
523 } while (read_seqretry(sl: &hh->hh_lock, start: seq));
524
525 if (WARN_ON_ONCE(skb_headroom(skb) < hh_alen)) {
526 kfree_skb(skb);
527 return NET_XMIT_DROP;
528 }
529
530 __skb_push(skb, len: hh_len);
531 return dev_queue_xmit(skb);
532}
533
534static inline int neigh_output(struct neighbour *n, struct sk_buff *skb,
535 bool skip_cache)
536{
537 const struct hh_cache *hh = &n->hh;
538
539 /* n->nud_state and hh->hh_len could be changed under us.
540 * neigh_hh_output() is taking care of the race later.
541 */
542 if (!skip_cache &&
543 (READ_ONCE(n->nud_state) & NUD_CONNECTED) &&
544 READ_ONCE(hh->hh_len))
545 return neigh_hh_output(hh, skb);
546
547 return READ_ONCE(n->output)(n, skb);
548}
549
550static inline struct neighbour *
551__neigh_lookup(struct neigh_table *tbl, const void *pkey, struct net_device *dev, int creat)
552{
553 struct neighbour *n = neigh_lookup(tbl, pkey, dev);
554
555 if (n || !creat)
556 return n;
557
558 n = neigh_create(tbl, pkey, dev);
559 return IS_ERR(ptr: n) ? NULL : n;
560}
561
562static inline struct neighbour *
563__neigh_lookup_errno(struct neigh_table *tbl, const void *pkey,
564 struct net_device *dev)
565{
566 struct neighbour *n = neigh_lookup(tbl, pkey, dev);
567
568 if (n)
569 return n;
570
571 return neigh_create(tbl, pkey, dev);
572}
573
574struct neighbour_cb {
575 unsigned long sched_next;
576 unsigned int flags;
577};
578
579#define LOCALLY_ENQUEUED 0x1
580
581#define NEIGH_CB(skb) ((struct neighbour_cb *)(skb)->cb)
582
583static inline void neigh_ha_snapshot(char *dst, const struct neighbour *n,
584 const struct net_device *dev)
585{
586 unsigned int seq;
587
588 do {
589 seq = read_seqbegin(sl: &n->ha_lock);
590 memcpy(to: dst, from: n->ha, len: dev->addr_len);
591 } while (read_seqretry(sl: &n->ha_lock, start: seq));
592}
593
594static inline void neigh_update_is_router(struct neighbour *neigh, u32 flags,
595 int *notify)
596{
597 u8 ndm_flags = 0;
598
599 ndm_flags |= (flags & NEIGH_UPDATE_F_ISROUTER) ? NTF_ROUTER : 0;
600 if ((neigh->flags ^ ndm_flags) & NTF_ROUTER) {
601 if (ndm_flags & NTF_ROUTER)
602 neigh->flags |= NTF_ROUTER;
603 else
604 neigh->flags &= ~NTF_ROUTER;
605 *notify = 1;
606 }
607}
608#endif
609