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