| 1 |
// SPDX-License-Identifier: GPL-2.0-or-later |
| 2 |
/* |
| 3 |
* Linux INET6 implementation |
| 4 |
* Forwarding Information Database |
| 5 |
* |
| 6 |
* Authors: |
| 7 |
* Pedro Roque <[email protected]> |
| 8 |
* |
| 9 |
* Changes: |
| 10 |
* Yuji SEKIYA @USAGI: Support default route on router node; |
| 11 |
* remove ip6_null_entry from the top of |
| 12 |
* routing table. |
| 13 |
* Ville Nuorvala: Fixed routing subtrees. |
| 14 |
*/ |
| 15 |
|
| 16 |
#define pr_fmt(fmt) "IPv6: " fmt |
| 17 |
|
| 18 |
#include <linux/bpf.h> |
| 19 |
#include <linux/errno.h> |
| 20 |
#include <linux/types.h> |
| 21 |
#include <linux/net.h> |
| 22 |
#include <linux/route.h> |
| 23 |
#include <linux/netdevice.h> |
| 24 |
#include <linux/in6.h> |
| 25 |
#include <linux/init.h> |
| 26 |
#include <linux/list.h> |
| 27 |
#include <linux/slab.h> |
| 28 |
|
| 29 |
#include <net/ip.h> |
| 30 |
#include <net/ipv6.h> |
| 31 |
#include <net/ndisc.h> |
| 32 |
#include <net/addrconf.h> |
| 33 |
#include <net/lwtunnel.h> |
| 34 |
#include <net/fib_notifier.h> |
| 35 |
|
| 36 |
#include <net/ip_fib.h> |
| 37 |
#include <net/ip6_fib.h> |
| 38 |
#include <net/ip6_route.h> |
| 39 |
|
| 40 |
static struct kmem_cache *fib6_node_kmem __read_mostly; |
| 41 |
|
| 42 |
struct fib6_cleaner { |
| 43 |
struct fib6_walker w; |
| 44 |
struct net *net; |
| 45 |
int (*func)(struct fib6_info *, void *arg); |
| 46 |
int sernum; |
| 47 |
void *arg; |
| 48 |
bool skip_notify; |
| 49 |
}; |
| 50 |
|
| 51 |
#ifdef CONFIG_IPV6_SUBTREES |
| 52 |
#define FWS_INIT FWS_S |
| 53 |
#else |
| 54 |
#define FWS_INIT FWS_L |
| 55 |
#endif |
| 56 |
|
| 57 |
static struct fib6_info *fib6_find_prefix(struct net *net, |
| 58 |
struct fib6_table *table, |
| 59 |
struct fib6_node *fn); |
| 60 |
static struct fib6_node *fib6_repair_tree(struct net *net, |
| 61 |
struct fib6_table *table, |
| 62 |
struct fib6_node *fn); |
| 63 |
static int fib6_walk(struct net *net, struct fib6_walker *w); |
| 64 |
static int fib6_walk_continue(struct fib6_walker *w); |
| 65 |
|
| 66 |
/* |
| 67 |
* A routing update causes an increase of the serial number on the |
| 68 |
* affected subtree. This allows for cached routes to be asynchronously |
| 69 |
* tested when modifications are made to the destination cache as a |
| 70 |
* result of redirects, path MTU changes, etc. |
| 71 |
*/ |
| 72 |
|
| 73 |
static void fib6_gc_timer_cb(struct timer_list *t); |
| 74 |
|
| 75 |
#define FOR_WALKERS(net, w) \ |
| 76 |
list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh) |
| 77 |
|
| 78 |
static void fib6_walker_link(struct net *net, struct fib6_walker *w) |
| 79 |
{ |
| 80 |
write_lock_bh(&net->ipv6.fib6_walker_lock); |
| 81 |
list_add(&w->lh, &net->ipv6.fib6_walkers); |
| 82 |
write_unlock_bh(&net->ipv6.fib6_walker_lock); |
| 83 |
} |
| 84 |
|
| 85 |
static void fib6_walker_unlink(struct net *net, struct fib6_walker *w) |
| 86 |
{ |
| 87 |
write_lock_bh(&net->ipv6.fib6_walker_lock); |
| 88 |
list_del(&w->lh); |
| 89 |
write_unlock_bh(&net->ipv6.fib6_walker_lock); |
| 90 |
} |
| 91 |
|
| 92 |
static int fib6_new_sernum(struct net *net) |
| 93 |
{ |
| 94 |
int new, old = atomic_read(&net->ipv6.fib6_sernum); |
| 95 |
|
| 96 |
do { |
| 97 |
new = old < INT_MAX ? old + 1 : 1; |
| 98 |
} while (!atomic_try_cmpxchg(&net->ipv6.fib6_sernum, &old, new)); |
| 99 |
|
| 100 |
return new; |
| 101 |
} |
| 102 |
|
| 103 |
enum { |
| 104 |
FIB6_NO_SERNUM_CHANGE = 0, |
| 105 |
}; |
| 106 |
|
| 107 |
void fib6_update_sernum(struct net *net, struct fib6_info *f6i) |
| 108 |
{ |
| 109 |
struct fib6_node *fn; |
| 110 |
|
| 111 |
fn = rcu_dereference_protected(f6i->fib6_node, |
| 112 |
lockdep_is_held(&f6i->fib6_table->tb6_lock)); |
| 113 |
if (fn) |
| 114 |
WRITE_ONCE(fn->fn_sernum, fib6_new_sernum(net)); |
| 115 |
} |
| 116 |
|
| 117 |
/* |
| 118 |
* Auxiliary address test functions for the radix tree. |
| 119 |
* |
| 120 |
* These assume a 32bit processor (although it will work on |
| 121 |
* 64bit processors) |
| 122 |
*/ |
| 123 |
|
| 124 |
/* |
| 125 |
* test bit |
| 126 |
*/ |
| 127 |
#if defined(__LITTLE_ENDIAN) |
| 128 |
# define BITOP_BE32_SWIZZLE (0x1F & ~7) |
| 129 |
#else |
| 130 |
# define BITOP_BE32_SWIZZLE 0 |
| 131 |
#endif |
| 132 |
|
| 133 |
static __be32 addr_bit_set(const void *token, int fn_bit) |
| 134 |
{ |
| 135 |
const __be32 *addr = token; |
| 136 |
/* |
| 137 |
* Here, |
| 138 |
* 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f) |
| 139 |
* is optimized version of |
| 140 |
* htonl(1 << ((~fn_bit)&0x1F)) |
| 141 |
* See include/asm-generic/bitops/le.h. |
| 142 |
*/ |
| 143 |
return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) & |
| 144 |
addr[fn_bit >> 5]; |
| 145 |
} |
| 146 |
|
| 147 |
struct fib6_info *fib6_info_alloc(gfp_t gfp_flags, bool with_fib6_nh) |
| 148 |
{ |
| 149 |
struct fib6_info *f6i; |
| 150 |
size_t sz = sizeof(*f6i); |
| 151 |
|
| 152 |
if (with_fib6_nh) |
| 153 |
sz += sizeof(struct fib6_nh); |
| 154 |
|
| 155 |
f6i = kzalloc(sz, gfp_flags); |
| 156 |
if (!f6i) |
| 157 |
return NULL; |
| 158 |
|
| 159 |
/* fib6_siblings is a union with nh_list, so this initializes both */ |
| 160 |
INIT_LIST_HEAD(&f6i->fib6_siblings); |
| 161 |
refcount_set(&f6i->fib6_ref, 1); |
| 162 |
|
| 163 |
INIT_HLIST_NODE(&f6i->gc_link); |
| 164 |
|
| 165 |
return f6i; |
| 166 |
} |
| 167 |
|
| 168 |
void fib6_info_destroy_rcu(struct rcu_head *head) |
| 169 |
{ |
| 170 |
struct fib6_info *f6i = container_of(head, struct fib6_info, rcu); |
| 171 |
|
| 172 |
WARN_ON(f6i->fib6_node); |
| 173 |
|
| 174 |
if (f6i->nh) |
| 175 |
nexthop_put(f6i->nh); |
| 176 |
else |
| 177 |
fib6_nh_release(f6i->fib6_nh); |
| 178 |
|
| 179 |
ip_fib_metrics_put(f6i->fib6_metrics); |
| 180 |
kfree(f6i); |
| 181 |
} |
| 182 |
EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu); |
| 183 |
|
| 184 |
static struct fib6_node *node_alloc(struct net *net) |
| 185 |
{ |
| 186 |
struct fib6_node *fn; |
| 187 |
|
| 188 |
fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC); |
| 189 |
if (fn) |
| 190 |
net->ipv6.rt6_stats->fib_nodes++; |
| 191 |
|
| 192 |
return fn; |
| 193 |
} |
| 194 |
|
| 195 |
static void node_free_immediate(struct net *net, struct fib6_node *fn) |
| 196 |
{ |
| 197 |
kmem_cache_free(fib6_node_kmem, fn); |
| 198 |
net->ipv6.rt6_stats->fib_nodes--; |
| 199 |
} |
| 200 |
|
| 201 |
static void node_free(struct net *net, struct fib6_node *fn) |
| 202 |
{ |
| 203 |
kfree_rcu(fn, rcu); |
| 204 |
net->ipv6.rt6_stats->fib_nodes--; |
| 205 |
} |
| 206 |
|
| 207 |
static void fib6_free_table(struct fib6_table *table) |
| 208 |
{ |
| 209 |
inetpeer_invalidate_tree(&table->tb6_peers); |
| 210 |
kfree(table); |
| 211 |
} |
| 212 |
|
| 213 |
static void fib6_link_table(struct net *net, struct fib6_table *tb) |
| 214 |
{ |
| 215 |
unsigned int h; |
| 216 |
|
| 217 |
/* |
| 218 |
* Initialize table lock at a single place to give lockdep a key, |
| 219 |
* tables aren't visible prior to being linked to the list. |
| 220 |
*/ |
| 221 |
spin_lock_init(&tb->tb6_lock); |
| 222 |
h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1); |
| 223 |
|
| 224 |
/* |
| 225 |
* No protection necessary, this is the only list mutatation |
| 226 |
* operation, tables never disappear once they exist. |
| 227 |
*/ |
| 228 |
hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]); |
| 229 |
} |
| 230 |
|
| 231 |
#ifdef CONFIG_IPV6_MULTIPLE_TABLES |
| 232 |
|
| 233 |
static struct fib6_table *fib6_alloc_table(struct net *net, u32 id) |
| 234 |
{ |
| 235 |
struct fib6_table *table; |
| 236 |
|
| 237 |
table = kzalloc_obj(*table, GFP_ATOMIC); |
| 238 |
if (table) { |
| 239 |
table->tb6_id = id; |
| 240 |
rcu_assign_pointer(table->tb6_root.leaf, |
| 241 |
net->ipv6.fib6_null_entry); |
| 242 |
table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; |
| 243 |
inet_peer_base_init(&table->tb6_peers); |
| 244 |
INIT_HLIST_HEAD(&table->tb6_gc_hlist); |
| 245 |
} |
| 246 |
|
| 247 |
return table; |
| 248 |
} |
| 249 |
|
| 250 |
struct fib6_table *fib6_new_table(struct net *net, u32 id) |
| 251 |
{ |
| 252 |
struct fib6_table *tb, *new_tb; |
| 253 |
|
| 254 |
if (id == 0) |
| 255 |
id = RT6_TABLE_MAIN; |
| 256 |
|
| 257 |
tb = fib6_get_table(net, id); |
| 258 |
if (tb) |
| 259 |
return tb; |
| 260 |
|
| 261 |
new_tb = fib6_alloc_table(net, id); |
| 262 |
if (!new_tb) |
| 263 |
return NULL; |
| 264 |
|
| 265 |
spin_lock_bh(&net->ipv6.fib_table_hash_lock); |
| 266 |
|
| 267 |
tb = fib6_get_table(net, id); |
| 268 |
if (unlikely(tb)) { |
| 269 |
spin_unlock_bh(&net->ipv6.fib_table_hash_lock); |
| 270 |
kfree(new_tb); |
| 271 |
return tb; |
| 272 |
} |
| 273 |
|
| 274 |
fib6_link_table(net, new_tb); |
| 275 |
|
| 276 |
spin_unlock_bh(&net->ipv6.fib_table_hash_lock); |
| 277 |
|
| 278 |
return new_tb; |
| 279 |
} |
| 280 |
EXPORT_SYMBOL_GPL(fib6_new_table); |
| 281 |
|
| 282 |
struct fib6_table *fib6_get_table(struct net *net, u32 id) |
| 283 |
{ |
| 284 |
struct hlist_head *head; |
| 285 |
struct fib6_table *tb; |
| 286 |
|
| 287 |
if (!id) |
| 288 |
id = RT6_TABLE_MAIN; |
| 289 |
|
| 290 |
head = &net->ipv6.fib_table_hash[id & (FIB6_TABLE_HASHSZ - 1)]; |
| 291 |
|
| 292 |
/* See comment in fib6_link_table(). RCU is not required, |
| 293 |
* but rcu_dereference_raw() is used to avoid data-race. |
| 294 |
*/ |
| 295 |
hlist_for_each_entry_rcu(tb, head, tb6_hlist, true) |
| 296 |
if (tb->tb6_id == id) |
| 297 |
return tb; |
| 298 |
|
| 299 |
return NULL; |
| 300 |
} |
| 301 |
EXPORT_SYMBOL_GPL(fib6_get_table); |
| 302 |
|
| 303 |
static void __net_init fib6_tables_init(struct net *net) |
| 304 |
{ |
| 305 |
fib6_link_table(net, net->ipv6.fib6_main_tbl); |
| 306 |
fib6_link_table(net, net->ipv6.fib6_local_tbl); |
| 307 |
} |
| 308 |
#else |
| 309 |
|
| 310 |
struct fib6_table *fib6_new_table(struct net *net, u32 id) |
| 311 |
{ |
| 312 |
return fib6_get_table(net, id); |
| 313 |
} |
| 314 |
|
| 315 |
struct fib6_table *fib6_get_table(struct net *net, u32 id) |
| 316 |
{ |
| 317 |
return net->ipv6.fib6_main_tbl; |
| 318 |
} |
| 319 |
|
| 320 |
struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6, |
| 321 |
const struct sk_buff *skb, |
| 322 |
int flags, pol_lookup_t lookup) |
| 323 |
{ |
| 324 |
struct rt6_info *rt; |
| 325 |
|
| 326 |
rt = pol_lookup_func(lookup, |
| 327 |
net, net->ipv6.fib6_main_tbl, fl6, skb, flags); |
| 328 |
if (rt->dst.error == -EAGAIN) { |
| 329 |
ip6_rt_put_flags(rt, flags); |
| 330 |
rt = net->ipv6.ip6_null_entry; |
| 331 |
if (!(flags & RT6_LOOKUP_F_DST_NOREF)) |
| 332 |
dst_hold(&rt->dst); |
| 333 |
} |
| 334 |
|
| 335 |
return &rt->dst; |
| 336 |
} |
| 337 |
|
| 338 |
/* called with rcu lock held; no reference taken on fib6_info */ |
| 339 |
int fib6_lookup(struct net *net, int oif, struct flowi6 *fl6, |
| 340 |
struct fib6_result *res, int flags) |
| 341 |
{ |
| 342 |
return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6, |
| 343 |
res, flags); |
| 344 |
} |
| 345 |
|
| 346 |
static void __net_init fib6_tables_init(struct net *net) |
| 347 |
{ |
| 348 |
fib6_link_table(net, net->ipv6.fib6_main_tbl); |
| 349 |
} |
| 350 |
|
| 351 |
#endif |
| 352 |
|
| 353 |
unsigned int fib6_tables_seq_read(const struct net *net) |
| 354 |
{ |
| 355 |
unsigned int h, fib_seq = 0; |
| 356 |
|
| 357 |
rcu_read_lock(); |
| 358 |
for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { |
| 359 |
const struct hlist_head *head = &net->ipv6.fib_table_hash[h]; |
| 360 |
const struct fib6_table *tb; |
| 361 |
|
| 362 |
hlist_for_each_entry_rcu(tb, head, tb6_hlist) |
| 363 |
fib_seq += READ_ONCE(tb->fib_seq); |
| 364 |
} |
| 365 |
rcu_read_unlock(); |
| 366 |
|
| 367 |
return fib_seq; |
| 368 |
} |
| 369 |
|
| 370 |
static int call_fib6_entry_notifier(struct notifier_block *nb, |
| 371 |
enum fib_event_type event_type, |
| 372 |
struct fib6_info *rt, |
| 373 |
struct netlink_ext_ack *extack) |
| 374 |
{ |
| 375 |
struct fib6_entry_notifier_info info = { |
| 376 |
.info.extack = extack, |
| 377 |
.rt = rt, |
| 378 |
}; |
| 379 |
|
| 380 |
return call_fib6_notifier(nb, event_type, &info.info); |
| 381 |
} |
| 382 |
|
| 383 |
static int call_fib6_multipath_entry_notifier(struct notifier_block *nb, |
| 384 |
enum fib_event_type event_type, |
| 385 |
struct fib6_info *rt, |
| 386 |
unsigned int nsiblings, |
| 387 |
struct netlink_ext_ack *extack) |
| 388 |
{ |
| 389 |
struct fib6_entry_notifier_info info = { |
| 390 |
.info.extack = extack, |
| 391 |
.rt = rt, |
| 392 |
.nsiblings = nsiblings, |
| 393 |
}; |
| 394 |
|
| 395 |
return call_fib6_notifier(nb, event_type, &info.info); |
| 396 |
} |
| 397 |
|
| 398 |
int call_fib6_entry_notifiers(struct net *net, |
| 399 |
enum fib_event_type event_type, |
| 400 |
struct fib6_info *rt, |
| 401 |
struct netlink_ext_ack *extack) |
| 402 |
{ |
| 403 |
struct fib6_entry_notifier_info info = { |
| 404 |
.info.extack = extack, |
| 405 |
.rt = rt, |
| 406 |
}; |
| 407 |
|
| 408 |
WRITE_ONCE(rt->fib6_table->fib_seq, rt->fib6_table->fib_seq + 1); |
| 409 |
return call_fib6_notifiers(net, event_type, &info.info); |
| 410 |
} |
| 411 |
|
| 412 |
int call_fib6_multipath_entry_notifiers(struct net *net, |
| 413 |
enum fib_event_type event_type, |
| 414 |
struct fib6_info *rt, |
| 415 |
unsigned int nsiblings, |
| 416 |
struct netlink_ext_ack *extack) |
| 417 |
{ |
| 418 |
struct fib6_entry_notifier_info info = { |
| 419 |
.info.extack = extack, |
| 420 |
.rt = rt, |
| 421 |
.nsiblings = nsiblings, |
| 422 |
}; |
| 423 |
|
| 424 |
WRITE_ONCE(rt->fib6_table->fib_seq, rt->fib6_table->fib_seq + 1); |
| 425 |
return call_fib6_notifiers(net, event_type, &info.info); |
| 426 |
} |
| 427 |
|
| 428 |
int call_fib6_entry_notifiers_replace(struct net *net, struct fib6_info *rt) |
| 429 |
{ |
| 430 |
struct fib6_entry_notifier_info info = { |
| 431 |
.rt = rt, |
| 432 |
.nsiblings = rt->fib6_nsiblings, |
| 433 |
}; |
| 434 |
|
| 435 |
WRITE_ONCE(rt->fib6_table->fib_seq, rt->fib6_table->fib_seq + 1); |
| 436 |
return call_fib6_notifiers(net, FIB_EVENT_ENTRY_REPLACE, &info.info); |
| 437 |
} |
| 438 |
|
| 439 |
struct fib6_dump_arg { |
| 440 |
struct net *net; |
| 441 |
struct notifier_block *nb; |
| 442 |
struct netlink_ext_ack *extack; |
| 443 |
}; |
| 444 |
|
| 445 |
static int fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg) |
| 446 |
{ |
| 447 |
enum fib_event_type fib_event = FIB_EVENT_ENTRY_REPLACE; |
| 448 |
unsigned int nsiblings; |
| 449 |
int err; |
| 450 |
|
| 451 |
if (!rt || rt == arg->net->ipv6.fib6_null_entry) |
| 452 |
return 0; |
| 453 |
|
| 454 |
nsiblings = READ_ONCE(rt->fib6_nsiblings); |
| 455 |
if (nsiblings) |
| 456 |
err = call_fib6_multipath_entry_notifier(arg->nb, fib_event, |
| 457 |
rt, |
| 458 |
nsiblings, |
| 459 |
arg->extack); |
| 460 |
else |
| 461 |
err = call_fib6_entry_notifier(arg->nb, fib_event, rt, |
| 462 |
arg->extack); |
| 463 |
|
| 464 |
return err; |
| 465 |
} |
| 466 |
|
| 467 |
static int fib6_node_dump(struct fib6_walker *w) |
| 468 |
{ |
| 469 |
int err; |
| 470 |
|
| 471 |
err = fib6_rt_dump(w->leaf, w->args); |
| 472 |
w->leaf = NULL; |
| 473 |
return err; |
| 474 |
} |
| 475 |
|
| 476 |
static int fib6_table_dump(struct net *net, struct fib6_table *tb, |
| 477 |
struct fib6_walker *w) |
| 478 |
{ |
| 479 |
int err; |
| 480 |
|
| 481 |
w->root = &tb->tb6_root; |
| 482 |
spin_lock_bh(&tb->tb6_lock); |
| 483 |
err = fib6_walk(net, w); |
| 484 |
spin_unlock_bh(&tb->tb6_lock); |
| 485 |
return err; |
| 486 |
} |
| 487 |
|
| 488 |
/* Called with rcu_read_lock() */ |
| 489 |
int fib6_tables_dump(struct net *net, struct notifier_block *nb, |
| 490 |
struct netlink_ext_ack *extack) |
| 491 |
{ |
| 492 |
struct fib6_dump_arg arg; |
| 493 |
struct fib6_walker *w; |
| 494 |
unsigned int h; |
| 495 |
int err = 0; |
| 496 |
|
| 497 |
w = kzalloc_obj(*w, GFP_ATOMIC); |
| 498 |
if (!w) |
| 499 |
return -ENOMEM; |
| 500 |
|
| 501 |
w->func = fib6_node_dump; |
| 502 |
arg.net = net; |
| 503 |
arg.nb = nb; |
| 504 |
arg.extack = extack; |
| 505 |
w->args = &arg; |
| 506 |
|
| 507 |
for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { |
| 508 |
struct hlist_head *head = &net->ipv6.fib_table_hash[h]; |
| 509 |
struct fib6_table *tb; |
| 510 |
|
| 511 |
hlist_for_each_entry_rcu(tb, head, tb6_hlist) { |
| 512 |
err = fib6_table_dump(net, tb, w); |
| 513 |
if (err) |
| 514 |
goto out; |
| 515 |
} |
| 516 |
} |
| 517 |
|
| 518 |
out: |
| 519 |
kfree(w); |
| 520 |
|
| 521 |
/* The tree traversal function should never return a positive value. */ |
| 522 |
return err > 0 ? -EINVAL : err; |
| 523 |
} |
| 524 |
|
| 525 |
static int fib6_dump_node(struct fib6_walker *w) |
| 526 |
{ |
| 527 |
int res; |
| 528 |
struct fib6_info *rt; |
| 529 |
|
| 530 |
for_each_fib6_walker_rt(w) { |
| 531 |
res = rt6_dump_route(rt, w->args, w->skip_in_node); |
| 532 |
if (res >= 0) { |
| 533 |
/* Frame is full, suspend walking */ |
| 534 |
w->leaf = rt; |
| 535 |
|
| 536 |
/* We'll restart from this node, so if some routes were |
| 537 |
* already dumped, skip them next time. |
| 538 |
*/ |
| 539 |
w->skip_in_node += res; |
| 540 |
|
| 541 |
return 1; |
| 542 |
} |
| 543 |
w->skip_in_node = 0; |
| 544 |
|
| 545 |
/* Multipath routes are dumped in one route with the |
| 546 |
* RTA_MULTIPATH attribute. Jump 'rt' to point to the |
| 547 |
* last sibling of this route (no need to dump the |
| 548 |
* sibling routes again) |
| 549 |
*/ |
| 550 |
if (rt->fib6_nsiblings) |
| 551 |
rt = list_last_entry(&rt->fib6_siblings, |
| 552 |
struct fib6_info, |
| 553 |
fib6_siblings); |
| 554 |
} |
| 555 |
w->leaf = NULL; |
| 556 |
return 0; |
| 557 |
} |
| 558 |
|
| 559 |
static void fib6_dump_end(struct netlink_callback *cb) |
| 560 |
{ |
| 561 |
struct net *net = sock_net(cb->skb->sk); |
| 562 |
struct fib6_walker *w = (void *)cb->args[2]; |
| 563 |
|
| 564 |
if (w) { |
| 565 |
if (cb->args[4]) { |
| 566 |
cb->args[4] = 0; |
| 567 |
fib6_walker_unlink(net, w); |
| 568 |
} |
| 569 |
cb->args[2] = 0; |
| 570 |
kfree(w); |
| 571 |
} |
| 572 |
cb->done = (void *)cb->args[3]; |
| 573 |
cb->args[1] = 3; |
| 574 |
} |
| 575 |
|
| 576 |
static int fib6_dump_done(struct netlink_callback *cb) |
| 577 |
{ |
| 578 |
fib6_dump_end(cb); |
| 579 |
return cb->done ? cb->done(cb) : 0; |
| 580 |
} |
| 581 |
|
| 582 |
static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb, |
| 583 |
struct netlink_callback *cb) |
| 584 |
{ |
| 585 |
struct net *net = sock_net(skb->sk); |
| 586 |
struct fib6_walker *w; |
| 587 |
int res; |
| 588 |
|
| 589 |
w = (void *)cb->args[2]; |
| 590 |
w->root = &table->tb6_root; |
| 591 |
|
| 592 |
if (cb->args[4] == 0) { |
| 593 |
w->count = 0; |
| 594 |
w->skip = 0; |
| 595 |
w->skip_in_node = 0; |
| 596 |
|
| 597 |
spin_lock_bh(&table->tb6_lock); |
| 598 |
res = fib6_walk(net, w); |
| 599 |
spin_unlock_bh(&table->tb6_lock); |
| 600 |
if (res > 0) { |
| 601 |
cb->args[4] = 1; |
| 602 |
cb->args[5] = READ_ONCE(w->root->fn_sernum); |
| 603 |
} |
| 604 |
} else { |
| 605 |
int sernum = READ_ONCE(w->root->fn_sernum); |
| 606 |
if (cb->args[5] != sernum) { |
| 607 |
/* Begin at the root if the tree changed */ |
| 608 |
cb->args[5] = sernum; |
| 609 |
w->state = FWS_INIT; |
| 610 |
w->node = w->root; |
| 611 |
w->skip = w->count; |
| 612 |
w->skip_in_node = 0; |
| 613 |
} else |
| 614 |
w->skip = 0; |
| 615 |
|
| 616 |
spin_lock_bh(&table->tb6_lock); |
| 617 |
res = fib6_walk_continue(w); |
| 618 |
spin_unlock_bh(&table->tb6_lock); |
| 619 |
if (res <= 0) { |
| 620 |
fib6_walker_unlink(net, w); |
| 621 |
cb->args[4] = 0; |
| 622 |
} |
| 623 |
} |
| 624 |
|
| 625 |
return res; |
| 626 |
} |
| 627 |
|
| 628 |
static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) |
| 629 |
{ |
| 630 |
struct rt6_rtnl_dump_arg arg = { |
| 631 |
.filter.dump_exceptions = true, |
| 632 |
.filter.dump_routes = true, |
| 633 |
.filter.rtnl_held = false, |
| 634 |
}; |
| 635 |
const struct nlmsghdr *nlh = cb->nlh; |
| 636 |
struct net *net = sock_net(skb->sk); |
| 637 |
unsigned int e = 0, s_e; |
| 638 |
struct hlist_head *head; |
| 639 |
struct fib6_walker *w; |
| 640 |
struct fib6_table *tb; |
| 641 |
unsigned int h, s_h; |
| 642 |
int err = 0; |
| 643 |
|
| 644 |
rcu_read_lock(); |
| 645 |
if (cb->strict_check) { |
| 646 |
err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb); |
| 647 |
if (err < 0) |
| 648 |
goto unlock; |
| 649 |
} else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) { |
| 650 |
struct rtmsg *rtm = nlmsg_data(nlh); |
| 651 |
|
| 652 |
if (rtm->rtm_flags & RTM_F_PREFIX) |
| 653 |
arg.filter.flags = RTM_F_PREFIX; |
| 654 |
} |
| 655 |
|
| 656 |
w = (void *)cb->args[2]; |
| 657 |
if (!w) { |
| 658 |
/* New dump: |
| 659 |
* |
| 660 |
* 1. allocate and initialize walker. |
| 661 |
*/ |
| 662 |
w = kzalloc_obj(*w, GFP_ATOMIC); |
| 663 |
if (!w) { |
| 664 |
err = -ENOMEM; |
| 665 |
goto unlock; |
| 666 |
} |
| 667 |
w->func = fib6_dump_node; |
| 668 |
cb->args[2] = (long)w; |
| 669 |
|
| 670 |
/* 2. hook callback destructor. |
| 671 |
*/ |
| 672 |
cb->args[3] = (long)cb->done; |
| 673 |
cb->done = fib6_dump_done; |
| 674 |
|
| 675 |
} |
| 676 |
|
| 677 |
arg.skb = skb; |
| 678 |
arg.cb = cb; |
| 679 |
arg.net = net; |
| 680 |
w->args = &arg; |
| 681 |
|
| 682 |
if (arg.filter.table_id) { |
| 683 |
tb = fib6_get_table(net, arg.filter.table_id); |
| 684 |
if (!tb) { |
| 685 |
if (rtnl_msg_family(cb->nlh) != PF_INET6) |
| 686 |
goto unlock; |
| 687 |
|
| 688 |
NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist"); |
| 689 |
err = -ENOENT; |
| 690 |
goto unlock; |
| 691 |
} |
| 692 |
|
| 693 |
if (!cb->args[0]) { |
| 694 |
err = fib6_dump_table(tb, skb, cb); |
| 695 |
if (!err) |
| 696 |
cb->args[0] = 1; |
| 697 |
} |
| 698 |
goto unlock; |
| 699 |
} |
| 700 |
|
| 701 |
s_h = cb->args[0]; |
| 702 |
s_e = cb->args[1]; |
| 703 |
|
| 704 |
for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) { |
| 705 |
e = 0; |
| 706 |
head = &net->ipv6.fib_table_hash[h]; |
| 707 |
hlist_for_each_entry_rcu(tb, head, tb6_hlist) { |
| 708 |
if (e < s_e) |
| 709 |
goto next; |
| 710 |
err = fib6_dump_table(tb, skb, cb); |
| 711 |
if (err != 0) |
| 712 |
goto out; |
| 713 |
next: |
| 714 |
e++; |
| 715 |
} |
| 716 |
} |
| 717 |
out: |
| 718 |
cb->args[1] = e; |
| 719 |
cb->args[0] = h; |
| 720 |
|
| 721 |
unlock: |
| 722 |
rcu_read_unlock(); |
| 723 |
if (err <= 0) |
| 724 |
fib6_dump_end(cb); |
| 725 |
return err; |
| 726 |
} |
| 727 |
|
| 728 |
void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val) |
| 729 |
{ |
| 730 |
struct dst_metrics *m; |
| 731 |
|
| 732 |
if (!f6i) |
| 733 |
return; |
| 734 |
|
| 735 |
if (READ_ONCE(f6i->fib6_metrics) == &dst_default_metrics) { |
| 736 |
struct dst_metrics *dflt = (struct dst_metrics *)&dst_default_metrics; |
| 737 |
struct dst_metrics *p = kzalloc_obj(*p, GFP_ATOMIC); |
| 738 |
|
| 739 |
if (!p) |
| 740 |
return; |
| 741 |
|
| 742 |
p->metrics[metric - 1] = val; |
| 743 |
refcount_set(&p->refcnt, 1); |
| 744 |
if (cmpxchg(&f6i->fib6_metrics, dflt, p) != dflt) |
| 745 |
kfree(p); |
| 746 |
else |
| 747 |
return; |
| 748 |
} |
| 749 |
|
| 750 |
m = READ_ONCE(f6i->fib6_metrics); |
| 751 |
WRITE_ONCE(m->metrics[metric - 1], val); |
| 752 |
} |
| 753 |
|
| 754 |
/* |
| 755 |
* Routing Table |
| 756 |
* |
| 757 |
* return the appropriate node for a routing tree "add" operation |
| 758 |
* by either creating and inserting or by returning an existing |
| 759 |
* node. |
| 760 |
*/ |
| 761 |
|
| 762 |
static struct fib6_node *fib6_add_1(struct net *net, |
| 763 |
struct fib6_table *table, |
| 764 |
struct fib6_node *root, |
| 765 |
struct in6_addr *addr, int plen, |
| 766 |
int offset, int allow_create, |
| 767 |
int replace_required, |
| 768 |
struct netlink_ext_ack *extack) |
| 769 |
{ |
| 770 |
struct fib6_node *fn, *in, *ln; |
| 771 |
struct fib6_node *pn = NULL; |
| 772 |
struct rt6key *key; |
| 773 |
int bit; |
| 774 |
__be32 dir = 0; |
| 775 |
|
| 776 |
/* insert node in tree */ |
| 777 |
|
| 778 |
fn = root; |
| 779 |
|
| 780 |
do { |
| 781 |
struct fib6_info *leaf = rcu_dereference_protected(fn->leaf, |
| 782 |
lockdep_is_held(&table->tb6_lock)); |
| 783 |
key = (struct rt6key *)((u8 *)leaf + offset); |
| 784 |
|
| 785 |
/* |
| 786 |
* Prefix match |
| 787 |
*/ |
| 788 |
if (plen < fn->fn_bit || |
| 789 |
!ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) { |
| 790 |
if (!allow_create) { |
| 791 |
if (replace_required) { |
| 792 |
NL_SET_ERR_MSG(extack, |
| 793 |
"Can not replace route - no match found"); |
| 794 |
pr_warn("Can't replace route, no match found\n"); |
| 795 |
return ERR_PTR(-ENOENT); |
| 796 |
} |
| 797 |
pr_warn("NLM_F_CREATE should be set when creating new route\n"); |
| 798 |
} |
| 799 |
goto insert_above; |
| 800 |
} |
| 801 |
|
| 802 |
/* |
| 803 |
* Exact match ? |
| 804 |
*/ |
| 805 |
|
| 806 |
if (plen == fn->fn_bit) { |
| 807 |
/* clean up an intermediate node */ |
| 808 |
if (!(fn->fn_flags & RTN_RTINFO)) { |
| 809 |
RCU_INIT_POINTER(fn->leaf, NULL); |
| 810 |
fib6_info_release(leaf); |
| 811 |
/* remove null_entry in the root node */ |
| 812 |
} else if (fn->fn_flags & RTN_TL_ROOT && |
| 813 |
rcu_access_pointer(fn->leaf) == |
| 814 |
net->ipv6.fib6_null_entry) { |
| 815 |
RCU_INIT_POINTER(fn->leaf, NULL); |
| 816 |
} |
| 817 |
|
| 818 |
return fn; |
| 819 |
} |
| 820 |
|
| 821 |
/* |
| 822 |
* We have more bits to go |
| 823 |
*/ |
| 824 |
|
| 825 |
/* Try to walk down on tree. */ |
| 826 |
dir = addr_bit_set(addr, fn->fn_bit); |
| 827 |
pn = fn; |
| 828 |
fn = dir ? |
| 829 |
rcu_dereference_protected(fn->right, |
| 830 |
lockdep_is_held(&table->tb6_lock)) : |
| 831 |
rcu_dereference_protected(fn->left, |
| 832 |
lockdep_is_held(&table->tb6_lock)); |
| 833 |
} while (fn); |
| 834 |
|
| 835 |
if (!allow_create) { |
| 836 |
/* We should not create new node because |
| 837 |
* NLM_F_REPLACE was specified without NLM_F_CREATE |
| 838 |
* I assume it is safe to require NLM_F_CREATE when |
| 839 |
* REPLACE flag is used! Later we may want to remove the |
| 840 |
* check for replace_required, because according |
| 841 |
* to netlink specification, NLM_F_CREATE |
| 842 |
* MUST be specified if new route is created. |
| 843 |
* That would keep IPv6 consistent with IPv4 |
| 844 |
*/ |
| 845 |
if (replace_required) { |
| 846 |
NL_SET_ERR_MSG(extack, |
| 847 |
"Can not replace route - no match found"); |
| 848 |
pr_warn("Can't replace route, no match found\n"); |
| 849 |
return ERR_PTR(-ENOENT); |
| 850 |
} |
| 851 |
pr_warn("NLM_F_CREATE should be set when creating new route\n"); |
| 852 |
} |
| 853 |
/* |
| 854 |
* We walked to the bottom of tree. |
| 855 |
* Create new leaf node without children. |
| 856 |
*/ |
| 857 |
|
| 858 |
ln = node_alloc(net); |
| 859 |
|
| 860 |
if (!ln) |
| 861 |
return ERR_PTR(-ENOMEM); |
| 862 |
ln->fn_bit = plen; |
| 863 |
RCU_INIT_POINTER(ln->parent, pn); |
| 864 |
|
| 865 |
if (dir) |
| 866 |
rcu_assign_pointer(pn->right, ln); |
| 867 |
else |
| 868 |
rcu_assign_pointer(pn->left, ln); |
| 869 |
|
| 870 |
return ln; |
| 871 |
|
| 872 |
|
| 873 |
insert_above: |
| 874 |
/* |
| 875 |
* split since we don't have a common prefix anymore or |
| 876 |
* we have a less significant route. |
| 877 |
* we've to insert an intermediate node on the list |
| 878 |
* this new node will point to the one we need to create |
| 879 |
* and the current |
| 880 |
*/ |
| 881 |
|
| 882 |
pn = rcu_dereference_protected(fn->parent, |
| 883 |
lockdep_is_held(&table->tb6_lock)); |
| 884 |
|
| 885 |
/* find 1st bit in difference between the 2 addrs. |
| 886 |
|
| 887 |
See comment in __ipv6_addr_diff: bit may be an invalid value, |
| 888 |
but if it is >= plen, the value is ignored in any case. |
| 889 |
*/ |
| 890 |
|
| 891 |
bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr)); |
| 892 |
|
| 893 |
/* |
| 894 |
* (intermediate)[in] |
| 895 |
* / \ |
| 896 |
* (new leaf node)[ln] (old node)[fn] |
| 897 |
*/ |
| 898 |
if (plen > bit) { |
| 899 |
in = node_alloc(net); |
| 900 |
ln = node_alloc(net); |
| 901 |
|
| 902 |
if (!in || !ln) { |
| 903 |
if (in) |
| 904 |
node_free_immediate(net, in); |
| 905 |
if (ln) |
| 906 |
node_free_immediate(net, ln); |
| 907 |
return ERR_PTR(-ENOMEM); |
| 908 |
} |
| 909 |
|
| 910 |
/* |
| 911 |
* new intermediate node. |
| 912 |
* RTN_RTINFO will |
| 913 |
* be off since that an address that chooses one of |
| 914 |
* the branches would not match less specific routes |
| 915 |
* in the other branch |
| 916 |
*/ |
| 917 |
|
| 918 |
in->fn_bit = bit; |
| 919 |
|
| 920 |
RCU_INIT_POINTER(in->parent, pn); |
| 921 |
in->leaf = fn->leaf; |
| 922 |
fib6_info_hold(rcu_dereference_protected(in->leaf, |
| 923 |
lockdep_is_held(&table->tb6_lock))); |
| 924 |
|
| 925 |
/* update parent pointer */ |
| 926 |
if (dir) |
| 927 |
rcu_assign_pointer(pn->right, in); |
| 928 |
else |
| 929 |
rcu_assign_pointer(pn->left, in); |
| 930 |
|
| 931 |
ln->fn_bit = plen; |
| 932 |
|
| 933 |
RCU_INIT_POINTER(ln->parent, in); |
| 934 |
rcu_assign_pointer(fn->parent, in); |
| 935 |
|
| 936 |
if (addr_bit_set(addr, bit)) { |
| 937 |
rcu_assign_pointer(in->right, ln); |
| 938 |
rcu_assign_pointer(in->left, fn); |
| 939 |
} else { |
| 940 |
rcu_assign_pointer(in->left, ln); |
| 941 |
rcu_assign_pointer(in->right, fn); |
| 942 |
} |
| 943 |
} else { /* plen <= bit */ |
| 944 |
|
| 945 |
/* |
| 946 |
* (new leaf node)[ln] |
| 947 |
* / \ |
| 948 |
* (old node)[fn] NULL |
| 949 |
*/ |
| 950 |
|
| 951 |
ln = node_alloc(net); |
| 952 |
|
| 953 |
if (!ln) |
| 954 |
return ERR_PTR(-ENOMEM); |
| 955 |
|
| 956 |
ln->fn_bit = plen; |
| 957 |
|
| 958 |
RCU_INIT_POINTER(ln->parent, pn); |
| 959 |
|
| 960 |
if (addr_bit_set(&key->addr, plen)) |
| 961 |
RCU_INIT_POINTER(ln->right, fn); |
| 962 |
else |
| 963 |
RCU_INIT_POINTER(ln->left, fn); |
| 964 |
|
| 965 |
rcu_assign_pointer(fn->parent, ln); |
| 966 |
|
| 967 |
if (dir) |
| 968 |
rcu_assign_pointer(pn->right, ln); |
| 969 |
else |
| 970 |
rcu_assign_pointer(pn->left, ln); |
| 971 |
} |
| 972 |
return ln; |
| 973 |
} |
| 974 |
|
| 975 |
static void __fib6_drop_pcpu_from(struct fib6_nh *fib6_nh, |
| 976 |
const struct fib6_info *match) |
| 977 |
{ |
| 978 |
int cpu; |
| 979 |
|
| 980 |
if (!fib6_nh->rt6i_pcpu) |
| 981 |
return; |
| 982 |
|
| 983 |
rcu_read_lock(); |
| 984 |
/* release the reference to this fib entry from |
| 985 |
* all of its cached pcpu routes |
| 986 |
*/ |
| 987 |
for_each_possible_cpu(cpu) { |
| 988 |
struct rt6_info **ppcpu_rt; |
| 989 |
struct rt6_info *pcpu_rt; |
| 990 |
|
| 991 |
ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu); |
| 992 |
|
| 993 |
/* Paired with xchg() in rt6_get_pcpu_route() */ |
| 994 |
pcpu_rt = READ_ONCE(*ppcpu_rt); |
| 995 |
|
| 996 |
/* only dropping the 'from' reference if the cached route |
| 997 |
* is using 'match'. The cached pcpu_rt->from only changes |
| 998 |
* from a fib6_info to NULL (ip6_dst_destroy); it can never |
| 999 |
* change from one fib6_info reference to another |
| 1000 |
*/ |
| 1001 |
if (pcpu_rt && rcu_access_pointer(pcpu_rt->from) == match) { |
| 1002 |
struct fib6_info *from; |
| 1003 |
|
| 1004 |
from = unrcu_pointer(xchg(&pcpu_rt->from, NULL)); |
| 1005 |
fib6_info_release(from); |
| 1006 |
} |
| 1007 |
} |
| 1008 |
rcu_read_unlock(); |
| 1009 |
} |
| 1010 |
|
| 1011 |
static int fib6_nh_drop_pcpu_from(struct fib6_nh *nh, void *_arg) |
| 1012 |
{ |
| 1013 |
struct fib6_info *arg = _arg; |
| 1014 |
|
| 1015 |
__fib6_drop_pcpu_from(nh, arg); |
| 1016 |
return 0; |
| 1017 |
} |
| 1018 |
|
| 1019 |
static void fib6_drop_pcpu_from(struct fib6_info *f6i) |
| 1020 |
{ |
| 1021 |
/* Make sure rt6_make_pcpu_route() wont add other percpu routes |
| 1022 |
* while we are cleaning them here. |
| 1023 |
*/ |
| 1024 |
f6i->fib6_destroying = 1; |
| 1025 |
mb(); /* paired with the cmpxchg() in rt6_make_pcpu_route() */ |
| 1026 |
|
| 1027 |
if (f6i->nh) { |
| 1028 |
rcu_read_lock(); |
| 1029 |
nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_drop_pcpu_from, f6i); |
| 1030 |
rcu_read_unlock(); |
| 1031 |
} else { |
| 1032 |
struct fib6_nh *fib6_nh; |
| 1033 |
|
| 1034 |
fib6_nh = f6i->fib6_nh; |
| 1035 |
__fib6_drop_pcpu_from(fib6_nh, f6i); |
| 1036 |
} |
| 1037 |
} |
| 1038 |
|
| 1039 |
static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn, |
| 1040 |
struct net *net) |
| 1041 |
{ |
| 1042 |
struct fib6_table *table = rt->fib6_table; |
| 1043 |
|
| 1044 |
/* Flush all cached dst in exception table */ |
| 1045 |
rt6_flush_exceptions(rt); |
| 1046 |
fib6_drop_pcpu_from(rt); |
| 1047 |
|
| 1048 |
if (rt->nh) { |
| 1049 |
spin_lock(&rt->nh->lock); |
| 1050 |
|
| 1051 |
if (!list_empty(&rt->nh_list)) |
| 1052 |
list_del_init(&rt->nh_list); |
| 1053 |
|
| 1054 |
spin_unlock(&rt->nh->lock); |
| 1055 |
} |
| 1056 |
|
| 1057 |
if (refcount_read(&rt->fib6_ref) != 1) { |
| 1058 |
/* This route is used as dummy address holder in some split |
| 1059 |
* nodes. It is not leaked, but it still holds other resources, |
| 1060 |
* which must be released in time. So, scan ascendant nodes |
| 1061 |
* and replace dummy references to this route with references |
| 1062 |
* to still alive ones. |
| 1063 |
*/ |
| 1064 |
while (fn) { |
| 1065 |
struct fib6_info *leaf = rcu_dereference_protected(fn->leaf, |
| 1066 |
lockdep_is_held(&table->tb6_lock)); |
| 1067 |
struct fib6_info *new_leaf; |
| 1068 |
if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) { |
| 1069 |
new_leaf = fib6_find_prefix(net, table, fn); |
| 1070 |
fib6_info_hold(new_leaf); |
| 1071 |
|
| 1072 |
rcu_assign_pointer(fn->leaf, new_leaf); |
| 1073 |
fib6_info_release(rt); |
| 1074 |
} |
| 1075 |
fn = rcu_dereference_protected(fn->parent, |
| 1076 |
lockdep_is_held(&table->tb6_lock)); |
| 1077 |
} |
| 1078 |
} |
| 1079 |
|
| 1080 |
fib6_clean_expires(rt); |
| 1081 |
fib6_remove_gc_list(rt); |
| 1082 |
} |
| 1083 |
|
| 1084 |
/* |
| 1085 |
* Insert routing information in a node. |
| 1086 |
*/ |
| 1087 |
|
| 1088 |
static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt, |
| 1089 |
struct nl_info *info, struct netlink_ext_ack *extack, |
| 1090 |
struct list_head *purge_list) |
| 1091 |
{ |
| 1092 |
struct fib6_info *leaf = rcu_dereference_protected(fn->leaf, |
| 1093 |
lockdep_is_held(&rt->fib6_table->tb6_lock)); |
| 1094 |
struct fib6_info *iter = NULL; |
| 1095 |
struct fib6_info __rcu **ins; |
| 1096 |
struct fib6_info __rcu **fallback_ins = NULL; |
| 1097 |
int replace = (info->nlh && |
| 1098 |
(info->nlh->nlmsg_flags & NLM_F_REPLACE)); |
| 1099 |
int add = (!info->nlh || |
| 1100 |
(info->nlh->nlmsg_flags & NLM_F_CREATE)); |
| 1101 |
int found = 0; |
| 1102 |
bool rt_can_ecmp = rt6_qualify_for_ecmp(rt); |
| 1103 |
bool notify_sibling_rt = false; |
| 1104 |
u16 nlflags = NLM_F_EXCL; |
| 1105 |
int err; |
| 1106 |
|
| 1107 |
if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND)) |
| 1108 |
nlflags |= NLM_F_APPEND; |
| 1109 |
|
| 1110 |
ins = &fn->leaf; |
| 1111 |
|
| 1112 |
for (iter = leaf; iter; |
| 1113 |
iter = rcu_dereference_protected(iter->fib6_next, |
| 1114 |
lockdep_is_held(&rt->fib6_table->tb6_lock))) { |
| 1115 |
/* |
| 1116 |
* Search for duplicates |
| 1117 |
*/ |
| 1118 |
|
| 1119 |
if (iter->fib6_metric == rt->fib6_metric) { |
| 1120 |
/* |
| 1121 |
* Same priority level |
| 1122 |
*/ |
| 1123 |
if (info->nlh && |
| 1124 |
(info->nlh->nlmsg_flags & NLM_F_EXCL)) |
| 1125 |
return -EEXIST; |
| 1126 |
|
| 1127 |
nlflags &= ~NLM_F_EXCL; |
| 1128 |
if (replace) { |
| 1129 |
if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) { |
| 1130 |
found++; |
| 1131 |
break; |
| 1132 |
} |
| 1133 |
fallback_ins = fallback_ins ?: ins; |
| 1134 |
goto next_iter; |
| 1135 |
} |
| 1136 |
|
| 1137 |
if (rt6_duplicate_nexthop(iter, rt)) { |
| 1138 |
if (rt->fib6_nsiblings) |
| 1139 |
WRITE_ONCE(rt->fib6_nsiblings, 0); |
| 1140 |
if (!(iter->fib6_flags & RTF_EXPIRES)) |
| 1141 |
return -EEXIST; |
| 1142 |
if (!(rt->fib6_flags & RTF_EXPIRES)) { |
| 1143 |
fib6_clean_expires(iter); |
| 1144 |
fib6_may_remove_gc_list(info->nl_net, iter); |
| 1145 |
} else { |
| 1146 |
fib6_set_expires(iter, rt->expires); |
| 1147 |
fib6_add_gc_list(iter); |
| 1148 |
} |
| 1149 |
if (!(rt->fib6_flags & (RTF_ADDRCONF | RTF_PREFIX_RT)) && |
| 1150 |
(iter->nh || !iter->fib6_nh->fib_nh_gw_family)) { |
| 1151 |
iter->fib6_flags &= ~RTF_ADDRCONF; |
| 1152 |
iter->fib6_flags &= ~RTF_PREFIX_RT; |
| 1153 |
} |
| 1154 |
|
| 1155 |
if (rt->fib6_pmtu) |
| 1156 |
fib6_metric_set(iter, RTAX_MTU, |
| 1157 |
rt->fib6_pmtu); |
| 1158 |
return -EEXIST; |
| 1159 |
} |
| 1160 |
/* If we have the same destination and the same metric, |
| 1161 |
* but not the same gateway, then the route we try to |
| 1162 |
* add is sibling to this route, increment our counter |
| 1163 |
* of siblings, and later we will add our route to the |
| 1164 |
* list. |
| 1165 |
* Only static routes (which don't have flag |
| 1166 |
* RTF_EXPIRES) are used for ECMPv6. |
| 1167 |
* |
| 1168 |
* To avoid long list, we only had siblings if the |
| 1169 |
* route have a gateway. |
| 1170 |
*/ |
| 1171 |
if (rt_can_ecmp && |
| 1172 |
rt6_qualify_for_ecmp(iter)) |
| 1173 |
WRITE_ONCE(rt->fib6_nsiblings, |
| 1174 |
rt->fib6_nsiblings + 1); |
| 1175 |
} |
| 1176 |
|
| 1177 |
if (iter->fib6_metric > rt->fib6_metric) |
| 1178 |
break; |
| 1179 |
|
| 1180 |
next_iter: |
| 1181 |
ins = &iter->fib6_next; |
| 1182 |
} |
| 1183 |
|
| 1184 |
if (fallback_ins && !found) { |
| 1185 |
/* No matching route with same ecmp-able-ness found, replace |
| 1186 |
* first matching route |
| 1187 |
*/ |
| 1188 |
ins = fallback_ins; |
| 1189 |
iter = rcu_dereference_protected(*ins, |
| 1190 |
lockdep_is_held(&rt->fib6_table->tb6_lock)); |
| 1191 |
found++; |
| 1192 |
} |
| 1193 |
|
| 1194 |
/* Reset round-robin state, if necessary */ |
| 1195 |
if (ins == &fn->leaf) |
| 1196 |
fn->rr_ptr = NULL; |
| 1197 |
|
| 1198 |
/* Link this route to others same route. */ |
| 1199 |
if (rt->fib6_nsiblings) { |
| 1200 |
unsigned int fib6_nsiblings; |
| 1201 |
struct fib6_info *sibling, *temp_sibling; |
| 1202 |
|
| 1203 |
/* Find the first route that have the same metric */ |
| 1204 |
sibling = leaf; |
| 1205 |
notify_sibling_rt = true; |
| 1206 |
while (sibling) { |
| 1207 |
if (sibling->fib6_metric == rt->fib6_metric && |
| 1208 |
rt6_qualify_for_ecmp(sibling)) { |
| 1209 |
list_add_tail_rcu(&rt->fib6_siblings, |
| 1210 |
&sibling->fib6_siblings); |
| 1211 |
break; |
| 1212 |
} |
| 1213 |
sibling = rcu_dereference_protected(sibling->fib6_next, |
| 1214 |
lockdep_is_held(&rt->fib6_table->tb6_lock)); |
| 1215 |
notify_sibling_rt = false; |
| 1216 |
} |
| 1217 |
/* For each sibling in the list, increment the counter of |
| 1218 |
* siblings. BUG() if counters does not match, list of siblings |
| 1219 |
* is broken! |
| 1220 |
*/ |
| 1221 |
fib6_nsiblings = 0; |
| 1222 |
list_for_each_entry_safe(sibling, temp_sibling, |
| 1223 |
&rt->fib6_siblings, fib6_siblings) { |
| 1224 |
WRITE_ONCE(sibling->fib6_nsiblings, |
| 1225 |
sibling->fib6_nsiblings + 1); |
| 1226 |
BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings); |
| 1227 |
fib6_nsiblings++; |
| 1228 |
} |
| 1229 |
BUG_ON(fib6_nsiblings != rt->fib6_nsiblings); |
| 1230 |
rcu_read_lock(); |
| 1231 |
rt6_multipath_rebalance(temp_sibling); |
| 1232 |
rcu_read_unlock(); |
| 1233 |
} |
| 1234 |
|
| 1235 |
/* |
| 1236 |
* insert node |
| 1237 |
*/ |
| 1238 |
if (!replace) { |
| 1239 |
if (!add) |
| 1240 |
pr_warn("NLM_F_CREATE should be set when creating new route\n"); |
| 1241 |
|
| 1242 |
add: |
| 1243 |
nlflags |= NLM_F_CREATE; |
| 1244 |
|
| 1245 |
/* The route should only be notified if it is the first |
| 1246 |
* route in the node or if it is added as a sibling |
| 1247 |
* route to the first route in the node. |
| 1248 |
*/ |
| 1249 |
if (!info->skip_notify_kernel && |
| 1250 |
(notify_sibling_rt || ins == &fn->leaf)) { |
| 1251 |
enum fib_event_type fib_event; |
| 1252 |
|
| 1253 |
if (notify_sibling_rt) |
| 1254 |
fib_event = FIB_EVENT_ENTRY_APPEND; |
| 1255 |
else |
| 1256 |
fib_event = FIB_EVENT_ENTRY_REPLACE; |
| 1257 |
err = call_fib6_entry_notifiers(info->nl_net, |
| 1258 |
fib_event, rt, |
| 1259 |
extack); |
| 1260 |
if (err) { |
| 1261 |
struct fib6_info *sibling, *next_sibling; |
| 1262 |
|
| 1263 |
/* If the route has siblings, then it first |
| 1264 |
* needs to be unlinked from them. |
| 1265 |
*/ |
| 1266 |
if (!rt->fib6_nsiblings) |
| 1267 |
return err; |
| 1268 |
|
| 1269 |
list_for_each_entry_safe(sibling, next_sibling, |
| 1270 |
&rt->fib6_siblings, |
| 1271 |
fib6_siblings) |
| 1272 |
WRITE_ONCE(sibling->fib6_nsiblings, |
| 1273 |
sibling->fib6_nsiblings - 1); |
| 1274 |
WRITE_ONCE(rt->fib6_nsiblings, 0); |
| 1275 |
list_del_rcu(&rt->fib6_siblings); |
| 1276 |
rcu_read_lock(); |
| 1277 |
rt6_multipath_rebalance(next_sibling); |
| 1278 |
rcu_read_unlock(); |
| 1279 |
return err; |
| 1280 |
} |
| 1281 |
} |
| 1282 |
|
| 1283 |
rcu_assign_pointer(rt->fib6_next, iter); |
| 1284 |
fib6_info_hold(rt); |
| 1285 |
rcu_assign_pointer(rt->fib6_node, fn); |
| 1286 |
rcu_assign_pointer(*ins, rt); |
| 1287 |
if (!info->skip_notify) |
| 1288 |
inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags); |
| 1289 |
info->nl_net->ipv6.rt6_stats->fib_rt_entries++; |
| 1290 |
|
| 1291 |
if (!(fn->fn_flags & RTN_RTINFO)) { |
| 1292 |
info->nl_net->ipv6.rt6_stats->fib_route_nodes++; |
| 1293 |
fn->fn_flags |= RTN_RTINFO; |
| 1294 |
} |
| 1295 |
|
| 1296 |
} else { |
| 1297 |
int nsiblings; |
| 1298 |
|
| 1299 |
if (!found) { |
| 1300 |
if (add) |
| 1301 |
goto add; |
| 1302 |
pr_warn("NLM_F_REPLACE set, but no existing node found!\n"); |
| 1303 |
return -ENOENT; |
| 1304 |
} |
| 1305 |
|
| 1306 |
if (!info->skip_notify_kernel && ins == &fn->leaf) { |
| 1307 |
err = call_fib6_entry_notifiers(info->nl_net, |
| 1308 |
FIB_EVENT_ENTRY_REPLACE, |
| 1309 |
rt, extack); |
| 1310 |
if (err) |
| 1311 |
return err; |
| 1312 |
} |
| 1313 |
|
| 1314 |
fib6_info_hold(rt); |
| 1315 |
rcu_assign_pointer(rt->fib6_node, fn); |
| 1316 |
rt->fib6_next = iter->fib6_next; |
| 1317 |
rcu_assign_pointer(*ins, rt); |
| 1318 |
if (!info->skip_notify) |
| 1319 |
inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE); |
| 1320 |
if (!(fn->fn_flags & RTN_RTINFO)) { |
| 1321 |
info->nl_net->ipv6.rt6_stats->fib_route_nodes++; |
| 1322 |
fn->fn_flags |= RTN_RTINFO; |
| 1323 |
} |
| 1324 |
nsiblings = iter->fib6_nsiblings; |
| 1325 |
iter->fib6_node = NULL; |
| 1326 |
list_add(&iter->purge_link, purge_list); |
| 1327 |
if (rcu_access_pointer(fn->rr_ptr) == iter) |
| 1328 |
fn->rr_ptr = NULL; |
| 1329 |
|
| 1330 |
if (nsiblings) { |
| 1331 |
/* Replacing an ECMP route, remove all siblings */ |
| 1332 |
ins = &rt->fib6_next; |
| 1333 |
iter = rcu_dereference_protected(*ins, |
| 1334 |
lockdep_is_held(&rt->fib6_table->tb6_lock)); |
| 1335 |
while (iter) { |
| 1336 |
if (iter->fib6_metric > rt->fib6_metric) |
| 1337 |
break; |
| 1338 |
if (rt6_qualify_for_ecmp(iter)) { |
| 1339 |
*ins = iter->fib6_next; |
| 1340 |
iter->fib6_node = NULL; |
| 1341 |
list_add(&iter->purge_link, purge_list); |
| 1342 |
if (rcu_access_pointer(fn->rr_ptr) == iter) |
| 1343 |
fn->rr_ptr = NULL; |
| 1344 |
nsiblings--; |
| 1345 |
info->nl_net->ipv6.rt6_stats->fib_rt_entries--; |
| 1346 |
} else { |
| 1347 |
ins = &iter->fib6_next; |
| 1348 |
} |
| 1349 |
iter = rcu_dereference_protected(*ins, |
| 1350 |
lockdep_is_held(&rt->fib6_table->tb6_lock)); |
| 1351 |
} |
| 1352 |
WARN_ON(nsiblings != 0); |
| 1353 |
} |
| 1354 |
} |
| 1355 |
|
| 1356 |
return 0; |
| 1357 |
} |
| 1358 |
|
| 1359 |
static int fib6_add_rt2node_nh(struct fib6_node *fn, struct fib6_info *rt, |
| 1360 |
struct nl_info *info, struct netlink_ext_ack *extack, |
| 1361 |
struct list_head *purge_list) |
| 1362 |
{ |
| 1363 |
int err; |
| 1364 |
|
| 1365 |
spin_lock(&rt->nh->lock); |
| 1366 |
|
| 1367 |
if (rt->nh->dead) { |
| 1368 |
NL_SET_ERR_MSG(extack, "Nexthop has been deleted"); |
| 1369 |
err = -EINVAL; |
| 1370 |
} else { |
| 1371 |
err = fib6_add_rt2node(fn, rt, info, extack, purge_list); |
| 1372 |
if (!err) |
| 1373 |
list_add(&rt->nh_list, &rt->nh->f6i_list); |
| 1374 |
} |
| 1375 |
|
| 1376 |
spin_unlock(&rt->nh->lock); |
| 1377 |
|
| 1378 |
return err; |
| 1379 |
} |
| 1380 |
|
| 1381 |
static void fib6_start_gc(struct net *net, struct fib6_info *rt) |
| 1382 |
{ |
| 1383 |
if (!timer_pending(&net->ipv6.ip6_fib_timer) && |
| 1384 |
(rt->fib6_flags & RTF_EXPIRES)) |
| 1385 |
mod_timer(&net->ipv6.ip6_fib_timer, |
| 1386 |
jiffies + READ_ONCE(net->ipv6.sysctl.ip6_rt_gc_interval)); |
| 1387 |
} |
| 1388 |
|
| 1389 |
void fib6_force_start_gc(struct net *net) |
| 1390 |
{ |
| 1391 |
if (!timer_pending(&net->ipv6.ip6_fib_timer)) |
| 1392 |
mod_timer(&net->ipv6.ip6_fib_timer, |
| 1393 |
jiffies + READ_ONCE(net->ipv6.sysctl.ip6_rt_gc_interval)); |
| 1394 |
} |
| 1395 |
|
| 1396 |
static void __fib6_update_sernum_upto_root(struct fib6_info *rt, |
| 1397 |
int sernum) |
| 1398 |
{ |
| 1399 |
struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node, |
| 1400 |
lockdep_is_held(&rt->fib6_table->tb6_lock)); |
| 1401 |
|
| 1402 |
/* paired with smp_rmb() in fib6_get_cookie_safe() */ |
| 1403 |
smp_wmb(); |
| 1404 |
while (fn) { |
| 1405 |
WRITE_ONCE(fn->fn_sernum, sernum); |
| 1406 |
fn = rcu_dereference_protected(fn->parent, |
| 1407 |
lockdep_is_held(&rt->fib6_table->tb6_lock)); |
| 1408 |
} |
| 1409 |
} |
| 1410 |
|
| 1411 |
void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt) |
| 1412 |
{ |
| 1413 |
__fib6_update_sernum_upto_root(rt, fib6_new_sernum(net)); |
| 1414 |
} |
| 1415 |
|
| 1416 |
/* allow ipv4 to update sernum via ipv6_stub */ |
| 1417 |
void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i) |
| 1418 |
{ |
| 1419 |
spin_lock_bh(&f6i->fib6_table->tb6_lock); |
| 1420 |
fib6_update_sernum_upto_root(net, f6i); |
| 1421 |
spin_unlock_bh(&f6i->fib6_table->tb6_lock); |
| 1422 |
} |
| 1423 |
|
| 1424 |
/* |
| 1425 |
* Add routing information to the routing tree. |
| 1426 |
* <destination addr>/<source addr> |
| 1427 |
* with source addr info in sub-trees |
| 1428 |
* Need to own table->tb6_lock |
| 1429 |
*/ |
| 1430 |
|
| 1431 |
int fib6_add(struct fib6_node *root, struct fib6_info *rt, |
| 1432 |
struct nl_info *info, struct netlink_ext_ack *extack) |
| 1433 |
{ |
| 1434 |
struct fib6_table *table = rt->fib6_table; |
| 1435 |
LIST_HEAD(purge_list); |
| 1436 |
struct fib6_node *fn; |
| 1437 |
#ifdef CONFIG_IPV6_SUBTREES |
| 1438 |
struct fib6_node *pn = NULL; |
| 1439 |
#endif |
| 1440 |
int err = -ENOMEM; |
| 1441 |
int allow_create = 1; |
| 1442 |
int replace_required = 0; |
| 1443 |
|
| 1444 |
if (info->nlh) { |
| 1445 |
if (!(info->nlh->nlmsg_flags & NLM_F_CREATE)) |
| 1446 |
allow_create = 0; |
| 1447 |
if (info->nlh->nlmsg_flags & NLM_F_REPLACE) |
| 1448 |
replace_required = 1; |
| 1449 |
} |
| 1450 |
if (!allow_create && !replace_required) |
| 1451 |
pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n"); |
| 1452 |
|
| 1453 |
fn = fib6_add_1(info->nl_net, table, root, |
| 1454 |
&rt->fib6_dst.addr, rt->fib6_dst.plen, |
| 1455 |
offsetof(struct fib6_info, fib6_dst), allow_create, |
| 1456 |
replace_required, extack); |
| 1457 |
if (IS_ERR(fn)) { |
| 1458 |
err = PTR_ERR(fn); |
| 1459 |
fn = NULL; |
| 1460 |
goto out; |
| 1461 |
} |
| 1462 |
|
| 1463 |
#ifdef CONFIG_IPV6_SUBTREES |
| 1464 |
pn = fn; |
| 1465 |
|
| 1466 |
if (rt->fib6_src.plen) { |
| 1467 |
struct fib6_node *sn; |
| 1468 |
|
| 1469 |
if (!rcu_access_pointer(fn->subtree)) { |
| 1470 |
struct fib6_node *sfn; |
| 1471 |
|
| 1472 |
/* |
| 1473 |
* Create subtree. |
| 1474 |
* |
| 1475 |
* fn[main tree] |
| 1476 |
* | |
| 1477 |
* sfn[subtree root] |
| 1478 |
* \ |
| 1479 |
* sn[new leaf node] |
| 1480 |
*/ |
| 1481 |
|
| 1482 |
/* Create subtree root node */ |
| 1483 |
sfn = node_alloc(info->nl_net); |
| 1484 |
if (!sfn) |
| 1485 |
goto failure; |
| 1486 |
|
| 1487 |
fib6_info_hold(info->nl_net->ipv6.fib6_null_entry); |
| 1488 |
rcu_assign_pointer(sfn->leaf, |
| 1489 |
info->nl_net->ipv6.fib6_null_entry); |
| 1490 |
sfn->fn_flags = RTN_ROOT; |
| 1491 |
|
| 1492 |
/* Now add the first leaf node to new subtree */ |
| 1493 |
|
| 1494 |
sn = fib6_add_1(info->nl_net, table, sfn, |
| 1495 |
&rt->fib6_src.addr, rt->fib6_src.plen, |
| 1496 |
offsetof(struct fib6_info, fib6_src), |
| 1497 |
allow_create, replace_required, extack); |
| 1498 |
|
| 1499 |
if (IS_ERR(sn)) { |
| 1500 |
/* If it is failed, discard just allocated |
| 1501 |
root, and then (in failure) stale node |
| 1502 |
in main tree. |
| 1503 |
*/ |
| 1504 |
node_free_immediate(info->nl_net, sfn); |
| 1505 |
err = PTR_ERR(sn); |
| 1506 |
goto failure; |
| 1507 |
} |
| 1508 |
|
| 1509 |
/* Now link new subtree to main tree */ |
| 1510 |
rcu_assign_pointer(sfn->parent, fn); |
| 1511 |
rcu_assign_pointer(fn->subtree, sfn); |
| 1512 |
} else { |
| 1513 |
sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn), |
| 1514 |
&rt->fib6_src.addr, rt->fib6_src.plen, |
| 1515 |
offsetof(struct fib6_info, fib6_src), |
| 1516 |
allow_create, replace_required, extack); |
| 1517 |
|
| 1518 |
if (IS_ERR(sn)) { |
| 1519 |
err = PTR_ERR(sn); |
| 1520 |
goto failure; |
| 1521 |
} |
| 1522 |
} |
| 1523 |
|
| 1524 |
if (!rcu_access_pointer(fn->leaf)) { |
| 1525 |
if (fn->fn_flags & RTN_TL_ROOT) { |
| 1526 |
/* put back null_entry for root node */ |
| 1527 |
rcu_assign_pointer(fn->leaf, |
| 1528 |
info->nl_net->ipv6.fib6_null_entry); |
| 1529 |
} else { |
| 1530 |
fib6_info_hold(rt); |
| 1531 |
rcu_assign_pointer(fn->leaf, rt); |
| 1532 |
} |
| 1533 |
} |
| 1534 |
fn = sn; |
| 1535 |
} |
| 1536 |
#endif |
| 1537 |
|
| 1538 |
if (rt->nh) |
| 1539 |
err = fib6_add_rt2node_nh(fn, rt, info, extack, &purge_list); |
| 1540 |
else |
| 1541 |
err = fib6_add_rt2node(fn, rt, info, extack, &purge_list); |
| 1542 |
if (!err) { |
| 1543 |
struct fib6_info *iter, *next; |
| 1544 |
|
| 1545 |
list_for_each_entry_safe(iter, next, &purge_list, purge_link) { |
| 1546 |
list_del(&iter->purge_link); |
| 1547 |
fib6_purge_rt(iter, fn, info->nl_net); |
| 1548 |
fib6_info_release(iter); |
| 1549 |
} |
| 1550 |
|
| 1551 |
__fib6_update_sernum_upto_root(rt, fib6_new_sernum(info->nl_net)); |
| 1552 |
|
| 1553 |
if (rt->fib6_flags & RTF_EXPIRES) |
| 1554 |
fib6_add_gc_list(rt); |
| 1555 |
|
| 1556 |
fib6_start_gc(info->nl_net, rt); |
| 1557 |
} |
| 1558 |
|
| 1559 |
out: |
| 1560 |
if (err) { |
| 1561 |
#ifdef CONFIG_IPV6_SUBTREES |
| 1562 |
/* |
| 1563 |
* If fib6_add_1 has cleared the old leaf pointer in the |
| 1564 |
* super-tree leaf node we have to find a new one for it. |
| 1565 |
*/ |
| 1566 |
if (pn != fn) { |
| 1567 |
struct fib6_info *pn_leaf = |
| 1568 |
rcu_dereference_protected(pn->leaf, |
| 1569 |
lockdep_is_held(&table->tb6_lock)); |
| 1570 |
if (pn_leaf == rt) { |
| 1571 |
pn_leaf = NULL; |
| 1572 |
RCU_INIT_POINTER(pn->leaf, NULL); |
| 1573 |
fib6_info_release(rt); |
| 1574 |
} |
| 1575 |
if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) { |
| 1576 |
pn_leaf = fib6_find_prefix(info->nl_net, table, |
| 1577 |
pn); |
| 1578 |
if (!pn_leaf) |
| 1579 |
pn_leaf = |
| 1580 |
info->nl_net->ipv6.fib6_null_entry; |
| 1581 |
fib6_info_hold(pn_leaf); |
| 1582 |
rcu_assign_pointer(pn->leaf, pn_leaf); |
| 1583 |
} |
| 1584 |
} |
| 1585 |
#endif |
| 1586 |
goto failure; |
| 1587 |
} else if (fib6_requires_src(rt)) { |
| 1588 |
fib6_routes_require_src_inc(info->nl_net); |
| 1589 |
} |
| 1590 |
return err; |
| 1591 |
|
| 1592 |
failure: |
| 1593 |
/* fn->leaf could be NULL and fib6_repair_tree() needs to be called if: |
| 1594 |
* 1. fn is an intermediate node and we failed to add the new |
| 1595 |
* route to it in both subtree creation failure and fib6_add_rt2node() |
| 1596 |
* failure case. |
| 1597 |
* 2. fn is the root node in the table and we fail to add the first |
| 1598 |
* default route to it. |
| 1599 |
*/ |
| 1600 |
if (fn && |
| 1601 |
(!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) || |
| 1602 |
(fn->fn_flags & RTN_TL_ROOT && |
| 1603 |
!rcu_access_pointer(fn->leaf)))) |
| 1604 |
fib6_repair_tree(info->nl_net, table, fn); |
| 1605 |
return err; |
| 1606 |
} |
| 1607 |
|
| 1608 |
/* |
| 1609 |
* Routing tree lookup |
| 1610 |
* |
| 1611 |
*/ |
| 1612 |
|
| 1613 |
struct lookup_args { |
| 1614 |
int offset; /* key offset on fib6_info */ |
| 1615 |
const struct in6_addr *addr; /* search key */ |
| 1616 |
}; |
| 1617 |
|
| 1618 |
static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root, |
| 1619 |
struct lookup_args *args) |
| 1620 |
{ |
| 1621 |
struct fib6_node *fn; |
| 1622 |
__be32 dir; |
| 1623 |
|
| 1624 |
if (unlikely(args->offset == 0)) |
| 1625 |
return NULL; |
| 1626 |
|
| 1627 |
/* |
| 1628 |
* Descend on a tree |
| 1629 |
*/ |
| 1630 |
|
| 1631 |
fn = root; |
| 1632 |
|
| 1633 |
for (;;) { |
| 1634 |
struct fib6_node *next; |
| 1635 |
|
| 1636 |
dir = addr_bit_set(args->addr, fn->fn_bit); |
| 1637 |
|
| 1638 |
next = dir ? rcu_dereference(fn->right) : |
| 1639 |
rcu_dereference(fn->left); |
| 1640 |
|
| 1641 |
if (next) { |
| 1642 |
fn = next; |
| 1643 |
continue; |
| 1644 |
} |
| 1645 |
break; |
| 1646 |
} |
| 1647 |
|
| 1648 |
while (fn) { |
| 1649 |
struct fib6_node *subtree = FIB6_SUBTREE(fn); |
| 1650 |
|
| 1651 |
if (subtree || fn->fn_flags & RTN_RTINFO) { |
| 1652 |
struct fib6_info *leaf = rcu_dereference(fn->leaf); |
| 1653 |
struct rt6key *key; |
| 1654 |
|
| 1655 |
if (!leaf) |
| 1656 |
goto backtrack; |
| 1657 |
|
| 1658 |
key = (struct rt6key *) ((u8 *)leaf + args->offset); |
| 1659 |
|
| 1660 |
if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) { |
| 1661 |
#ifdef CONFIG_IPV6_SUBTREES |
| 1662 |
if (subtree) { |
| 1663 |
struct fib6_node *sfn; |
| 1664 |
sfn = fib6_node_lookup_1(subtree, |
| 1665 |
args + 1); |
| 1666 |
if (!sfn) |
| 1667 |
goto backtrack; |
| 1668 |
fn = sfn; |
| 1669 |
} |
| 1670 |
#endif |
| 1671 |
if (fn->fn_flags & RTN_RTINFO) |
| 1672 |
return fn; |
| 1673 |
} |
| 1674 |
} |
| 1675 |
backtrack: |
| 1676 |
if (fn->fn_flags & RTN_ROOT) |
| 1677 |
break; |
| 1678 |
|
| 1679 |
fn = rcu_dereference(fn->parent); |
| 1680 |
} |
| 1681 |
|
| 1682 |
return NULL; |
| 1683 |
} |
| 1684 |
|
| 1685 |
/* called with rcu_read_lock() held |
| 1686 |
*/ |
| 1687 |
struct fib6_node *fib6_node_lookup(struct fib6_node *root, |
| 1688 |
const struct in6_addr *daddr, |
| 1689 |
const struct in6_addr *saddr) |
| 1690 |
{ |
| 1691 |
struct fib6_node *fn; |
| 1692 |
struct lookup_args args[] = { |
| 1693 |
{ |
| 1694 |
.offset = offsetof(struct fib6_info, fib6_dst), |
| 1695 |
.addr = daddr, |
| 1696 |
}, |
| 1697 |
#ifdef CONFIG_IPV6_SUBTREES |
| 1698 |
{ |
| 1699 |
.offset = offsetof(struct fib6_info, fib6_src), |
| 1700 |
.addr = saddr, |
| 1701 |
}, |
| 1702 |
#endif |
| 1703 |
{ |
| 1704 |
.offset = 0, /* sentinel */ |
| 1705 |
} |
| 1706 |
}; |
| 1707 |
|
| 1708 |
fn = fib6_node_lookup_1(root, daddr ? args : args + 1); |
| 1709 |
if (!fn || fn->fn_flags & RTN_TL_ROOT) |
| 1710 |
fn = root; |
| 1711 |
|
| 1712 |
return fn; |
| 1713 |
} |
| 1714 |
|
| 1715 |
/* |
| 1716 |
* Get node with specified destination prefix (and source prefix, |
| 1717 |
* if subtrees are used) |
| 1718 |
* exact_match == true means we try to find fn with exact match of |
| 1719 |
* the passed in prefix addr |
| 1720 |
* exact_match == false means we try to find fn with longest prefix |
| 1721 |
* match of the passed in prefix addr. This is useful for finding fn |
| 1722 |
* for cached route as it will be stored in the exception table under |
| 1723 |
* the node with longest prefix length. |
| 1724 |
*/ |
| 1725 |
|
| 1726 |
|
| 1727 |
static struct fib6_node *fib6_locate_1(struct fib6_node *root, |
| 1728 |
const struct in6_addr *addr, |
| 1729 |
int plen, int offset, |
| 1730 |
bool exact_match) |
| 1731 |
{ |
| 1732 |
struct fib6_node *fn, *prev = NULL; |
| 1733 |
|
| 1734 |
for (fn = root; fn ; ) { |
| 1735 |
struct fib6_info *leaf = rcu_dereference(fn->leaf); |
| 1736 |
struct rt6key *key; |
| 1737 |
|
| 1738 |
/* This node is being deleted */ |
| 1739 |
if (!leaf) { |
| 1740 |
if (plen <= fn->fn_bit) |
| 1741 |
goto out; |
| 1742 |
else |
| 1743 |
goto next; |
| 1744 |
} |
| 1745 |
|
| 1746 |
key = (struct rt6key *)((u8 *)leaf + offset); |
| 1747 |
|
| 1748 |
/* |
| 1749 |
* Prefix match |
| 1750 |
*/ |
| 1751 |
if (plen < fn->fn_bit || |
| 1752 |
!ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) |
| 1753 |
goto out; |
| 1754 |
|
| 1755 |
if (plen == fn->fn_bit) |
| 1756 |
return fn; |
| 1757 |
|
| 1758 |
if (fn->fn_flags & RTN_RTINFO) |
| 1759 |
prev = fn; |
| 1760 |
|
| 1761 |
next: |
| 1762 |
/* |
| 1763 |
* We have more bits to go |
| 1764 |
*/ |
| 1765 |
if (addr_bit_set(addr, fn->fn_bit)) |
| 1766 |
fn = rcu_dereference(fn->right); |
| 1767 |
else |
| 1768 |
fn = rcu_dereference(fn->left); |
| 1769 |
} |
| 1770 |
out: |
| 1771 |
if (exact_match) |
| 1772 |
return NULL; |
| 1773 |
else |
| 1774 |
return prev; |
| 1775 |
} |
| 1776 |
|
| 1777 |
struct fib6_node *fib6_locate(struct fib6_node *root, |
| 1778 |
const struct in6_addr *daddr, int dst_len, |
| 1779 |
const struct in6_addr *saddr, int src_len, |
| 1780 |
bool exact_match) |
| 1781 |
{ |
| 1782 |
struct fib6_node *fn; |
| 1783 |
|
| 1784 |
fn = fib6_locate_1(root, daddr, dst_len, |
| 1785 |
offsetof(struct fib6_info, fib6_dst), |
| 1786 |
exact_match); |
| 1787 |
|
| 1788 |
#ifdef CONFIG_IPV6_SUBTREES |
| 1789 |
if (src_len) { |
| 1790 |
WARN_ON(saddr == NULL); |
| 1791 |
if (fn) { |
| 1792 |
struct fib6_node *subtree = FIB6_SUBTREE(fn); |
| 1793 |
|
| 1794 |
if (subtree) { |
| 1795 |
fn = fib6_locate_1(subtree, saddr, src_len, |
| 1796 |
offsetof(struct fib6_info, fib6_src), |
| 1797 |
exact_match); |
| 1798 |
} |
| 1799 |
} |
| 1800 |
} |
| 1801 |
#endif |
| 1802 |
|
| 1803 |
if (fn && fn->fn_flags & RTN_RTINFO) |
| 1804 |
return fn; |
| 1805 |
|
| 1806 |
return NULL; |
| 1807 |
} |
| 1808 |
|
| 1809 |
|
| 1810 |
/* |
| 1811 |
* Deletion |
| 1812 |
* |
| 1813 |
*/ |
| 1814 |
|
| 1815 |
static struct fib6_info *fib6_find_prefix(struct net *net, |
| 1816 |
struct fib6_table *table, |
| 1817 |
struct fib6_node *fn) |
| 1818 |
{ |
| 1819 |
struct fib6_node *child_left, *child_right; |
| 1820 |
|
| 1821 |
if (fn->fn_flags & RTN_ROOT) |
| 1822 |
return net->ipv6.fib6_null_entry; |
| 1823 |
|
| 1824 |
while (fn) { |
| 1825 |
child_left = rcu_dereference_protected(fn->left, |
| 1826 |
lockdep_is_held(&table->tb6_lock)); |
| 1827 |
child_right = rcu_dereference_protected(fn->right, |
| 1828 |
lockdep_is_held(&table->tb6_lock)); |
| 1829 |
if (child_left) |
| 1830 |
return rcu_dereference_protected(child_left->leaf, |
| 1831 |
lockdep_is_held(&table->tb6_lock)); |
| 1832 |
if (child_right) |
| 1833 |
return rcu_dereference_protected(child_right->leaf, |
| 1834 |
lockdep_is_held(&table->tb6_lock)); |
| 1835 |
|
| 1836 |
fn = FIB6_SUBTREE(fn); |
| 1837 |
} |
| 1838 |
return NULL; |
| 1839 |
} |
| 1840 |
|
| 1841 |
/* |
| 1842 |
* Called to trim the tree of intermediate nodes when possible. "fn" |
| 1843 |
* is the node we want to try and remove. |
| 1844 |
* Need to own table->tb6_lock |
| 1845 |
*/ |
| 1846 |
|
| 1847 |
static struct fib6_node *fib6_repair_tree(struct net *net, |
| 1848 |
struct fib6_table *table, |
| 1849 |
struct fib6_node *fn) |
| 1850 |
{ |
| 1851 |
int children; |
| 1852 |
int nstate; |
| 1853 |
struct fib6_node *child; |
| 1854 |
struct fib6_walker *w; |
| 1855 |
int iter = 0; |
| 1856 |
|
| 1857 |
/* Set fn->leaf to null_entry for root node. */ |
| 1858 |
if (fn->fn_flags & RTN_TL_ROOT) { |
| 1859 |
rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry); |
| 1860 |
return fn; |
| 1861 |
} |
| 1862 |
|
| 1863 |
for (;;) { |
| 1864 |
struct fib6_node *fn_r = rcu_dereference_protected(fn->right, |
| 1865 |
lockdep_is_held(&table->tb6_lock)); |
| 1866 |
struct fib6_node *fn_l = rcu_dereference_protected(fn->left, |
| 1867 |
lockdep_is_held(&table->tb6_lock)); |
| 1868 |
struct fib6_node *pn = rcu_dereference_protected(fn->parent, |
| 1869 |
lockdep_is_held(&table->tb6_lock)); |
| 1870 |
struct fib6_node *pn_r = rcu_dereference_protected(pn->right, |
| 1871 |
lockdep_is_held(&table->tb6_lock)); |
| 1872 |
struct fib6_node *pn_l = rcu_dereference_protected(pn->left, |
| 1873 |
lockdep_is_held(&table->tb6_lock)); |
| 1874 |
struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf, |
| 1875 |
lockdep_is_held(&table->tb6_lock)); |
| 1876 |
struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf, |
| 1877 |
lockdep_is_held(&table->tb6_lock)); |
| 1878 |
struct fib6_info *new_fn_leaf; |
| 1879 |
|
| 1880 |
pr_debug("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter); |
| 1881 |
iter++; |
| 1882 |
|
| 1883 |
WARN_ON(fn->fn_flags & RTN_RTINFO); |
| 1884 |
WARN_ON(fn->fn_flags & RTN_TL_ROOT); |
| 1885 |
WARN_ON(fn_leaf); |
| 1886 |
|
| 1887 |
children = 0; |
| 1888 |
child = NULL; |
| 1889 |
if (fn_r) { |
| 1890 |
child = fn_r; |
| 1891 |
children |= 1; |
| 1892 |
} |
| 1893 |
if (fn_l) { |
| 1894 |
child = fn_l; |
| 1895 |
children |= 2; |
| 1896 |
} |
| 1897 |
|
| 1898 |
if (children == 3 || FIB6_SUBTREE(fn) |
| 1899 |
#ifdef CONFIG_IPV6_SUBTREES |
| 1900 |
/* Subtree root (i.e. fn) may have one child */ |
| 1901 |
|| (children && fn->fn_flags & RTN_ROOT) |
| 1902 |
#endif |
| 1903 |
) { |
| 1904 |
new_fn_leaf = fib6_find_prefix(net, table, fn); |
| 1905 |
#if RT6_DEBUG >= 2 |
| 1906 |
if (!new_fn_leaf) { |
| 1907 |
WARN_ON(!new_fn_leaf); |
| 1908 |
new_fn_leaf = net->ipv6.fib6_null_entry; |
| 1909 |
} |
| 1910 |
#endif |
| 1911 |
fib6_info_hold(new_fn_leaf); |
| 1912 |
rcu_assign_pointer(fn->leaf, new_fn_leaf); |
| 1913 |
return pn; |
| 1914 |
} |
| 1915 |
|
| 1916 |
#ifdef CONFIG_IPV6_SUBTREES |
| 1917 |
if (FIB6_SUBTREE(pn) == fn) { |
| 1918 |
WARN_ON(!(fn->fn_flags & RTN_ROOT)); |
| 1919 |
RCU_INIT_POINTER(pn->subtree, NULL); |
| 1920 |
nstate = FWS_L; |
| 1921 |
} else { |
| 1922 |
WARN_ON(fn->fn_flags & RTN_ROOT); |
| 1923 |
#endif |
| 1924 |
if (pn_r == fn) |
| 1925 |
rcu_assign_pointer(pn->right, child); |
| 1926 |
else if (pn_l == fn) |
| 1927 |
rcu_assign_pointer(pn->left, child); |
| 1928 |
#if RT6_DEBUG >= 2 |
| 1929 |
else |
| 1930 |
WARN_ON(1); |
| 1931 |
#endif |
| 1932 |
if (child) |
| 1933 |
rcu_assign_pointer(child->parent, pn); |
| 1934 |
nstate = FWS_R; |
| 1935 |
#ifdef CONFIG_IPV6_SUBTREES |
| 1936 |
} |
| 1937 |
#endif |
| 1938 |
|
| 1939 |
read_lock(&net->ipv6.fib6_walker_lock); |
| 1940 |
FOR_WALKERS(net, w) { |
| 1941 |
if (!child) { |
| 1942 |
if (w->node == fn) { |
| 1943 |
pr_debug("W %p adjusted by delnode 1, s=%d/%d\n", |
| 1944 |
w, w->state, nstate); |
| 1945 |
w->node = pn; |
| 1946 |
w->state = nstate; |
| 1947 |
} |
| 1948 |
} else { |
| 1949 |
if (w->node == fn) { |
| 1950 |
w->node = child; |
| 1951 |
if (children&2) { |
| 1952 |
pr_debug("W %p adjusted by delnode 2, s=%d\n", |
| 1953 |
w, w->state); |
| 1954 |
w->state = w->state >= FWS_R ? FWS_U : FWS_INIT; |
| 1955 |
} else { |
| 1956 |
pr_debug("W %p adjusted by delnode 2, s=%d\n", |
| 1957 |
w, w->state); |
| 1958 |
w->state = w->state >= FWS_C ? FWS_U : FWS_INIT; |
| 1959 |
} |
| 1960 |
} |
| 1961 |
} |
| 1962 |
} |
| 1963 |
read_unlock(&net->ipv6.fib6_walker_lock); |
| 1964 |
|
| 1965 |
node_free(net, fn); |
| 1966 |
if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn)) |
| 1967 |
return pn; |
| 1968 |
|
| 1969 |
RCU_INIT_POINTER(pn->leaf, NULL); |
| 1970 |
fib6_info_release(pn_leaf); |
| 1971 |
fn = pn; |
| 1972 |
} |
| 1973 |
} |
| 1974 |
|
| 1975 |
static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn, |
| 1976 |
struct fib6_info __rcu **rtp, struct nl_info *info) |
| 1977 |
{ |
| 1978 |
struct fib6_info *leaf, *replace_rt = NULL; |
| 1979 |
struct fib6_walker *w; |
| 1980 |
struct fib6_info *rt = rcu_dereference_protected(*rtp, |
| 1981 |
lockdep_is_held(&table->tb6_lock)); |
| 1982 |
struct net *net = info->nl_net; |
| 1983 |
bool notify_del = false; |
| 1984 |
|
| 1985 |
/* If the deleted route is the first in the node and it is not part of |
| 1986 |
* a multipath route, then we need to replace it with the next route |
| 1987 |
* in the node, if exists. |
| 1988 |
*/ |
| 1989 |
leaf = rcu_dereference_protected(fn->leaf, |
| 1990 |
lockdep_is_held(&table->tb6_lock)); |
| 1991 |
if (leaf == rt && !rt->fib6_nsiblings) { |
| 1992 |
if (rcu_access_pointer(rt->fib6_next)) |
| 1993 |
replace_rt = rcu_dereference_protected(rt->fib6_next, |
| 1994 |
lockdep_is_held(&table->tb6_lock)); |
| 1995 |
else |
| 1996 |
notify_del = true; |
| 1997 |
} |
| 1998 |
|
| 1999 |
/* Unlink it */ |
| 2000 |
*rtp = rt->fib6_next; |
| 2001 |
rt->fib6_node = NULL; |
| 2002 |
net->ipv6.rt6_stats->fib_rt_entries--; |
| 2003 |
net->ipv6.rt6_stats->fib_discarded_routes++; |
| 2004 |
|
| 2005 |
/* Reset round-robin state, if necessary */ |
| 2006 |
if (rcu_access_pointer(fn->rr_ptr) == rt) |
| 2007 |
fn->rr_ptr = NULL; |
| 2008 |
|
| 2009 |
/* Remove this entry from other siblings */ |
| 2010 |
if (rt->fib6_nsiblings) { |
| 2011 |
struct fib6_info *sibling, *next_sibling; |
| 2012 |
|
| 2013 |
/* The route is deleted from a multipath route. If this |
| 2014 |
* multipath route is the first route in the node, then we need |
| 2015 |
* to emit a delete notification. Otherwise, we need to skip |
| 2016 |
* the notification. |
| 2017 |
*/ |
| 2018 |
if (rt->fib6_metric == leaf->fib6_metric && |
| 2019 |
rt6_qualify_for_ecmp(leaf)) |
| 2020 |
notify_del = true; |
| 2021 |
list_for_each_entry_safe(sibling, next_sibling, |
| 2022 |
&rt->fib6_siblings, fib6_siblings) |
| 2023 |
WRITE_ONCE(sibling->fib6_nsiblings, |
| 2024 |
sibling->fib6_nsiblings - 1); |
| 2025 |
WRITE_ONCE(rt->fib6_nsiblings, 0); |
| 2026 |
list_del_rcu(&rt->fib6_siblings); |
| 2027 |
rt6_multipath_rebalance(next_sibling); |
| 2028 |
} |
| 2029 |
|
| 2030 |
/* Adjust walkers */ |
| 2031 |
read_lock(&net->ipv6.fib6_walker_lock); |
| 2032 |
FOR_WALKERS(net, w) { |
| 2033 |
if (w->state == FWS_C && w->leaf == rt) { |
| 2034 |
pr_debug("walker %p adjusted by delroute\n", w); |
| 2035 |
w->leaf = rcu_dereference_protected(rt->fib6_next, |
| 2036 |
lockdep_is_held(&table->tb6_lock)); |
| 2037 |
if (!w->leaf) |
| 2038 |
w->state = FWS_U; |
| 2039 |
} |
| 2040 |
} |
| 2041 |
read_unlock(&net->ipv6.fib6_walker_lock); |
| 2042 |
|
| 2043 |
/* If it was last route, call fib6_repair_tree() to: |
| 2044 |
* 1. For root node, put back null_entry as how the table was created. |
| 2045 |
* 2. For other nodes, expunge its radix tree node. |
| 2046 |
*/ |
| 2047 |
if (!rcu_access_pointer(fn->leaf)) { |
| 2048 |
if (!(fn->fn_flags & RTN_TL_ROOT)) { |
| 2049 |
fn->fn_flags &= ~RTN_RTINFO; |
| 2050 |
net->ipv6.rt6_stats->fib_route_nodes--; |
| 2051 |
} |
| 2052 |
fn = fib6_repair_tree(net, table, fn); |
| 2053 |
} |
| 2054 |
|
| 2055 |
fib6_purge_rt(rt, fn, net); |
| 2056 |
|
| 2057 |
if (!info->skip_notify_kernel) { |
| 2058 |
if (notify_del) |
| 2059 |
call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, |
| 2060 |
rt, NULL); |
| 2061 |
else if (replace_rt) |
| 2062 |
call_fib6_entry_notifiers_replace(net, replace_rt); |
| 2063 |
} |
| 2064 |
if (!info->skip_notify) |
| 2065 |
inet6_rt_notify(RTM_DELROUTE, rt, info, 0); |
| 2066 |
|
| 2067 |
fib6_info_release(rt); |
| 2068 |
} |
| 2069 |
|
| 2070 |
/* Need to own table->tb6_lock */ |
| 2071 |
int fib6_del(struct fib6_info *rt, struct nl_info *info) |
| 2072 |
{ |
| 2073 |
struct net *net = info->nl_net; |
| 2074 |
struct fib6_info __rcu **rtp; |
| 2075 |
struct fib6_info __rcu **rtp_next; |
| 2076 |
struct fib6_table *table; |
| 2077 |
struct fib6_node *fn; |
| 2078 |
|
| 2079 |
if (rt == net->ipv6.fib6_null_entry) |
| 2080 |
return -ENOENT; |
| 2081 |
|
| 2082 |
table = rt->fib6_table; |
| 2083 |
fn = rcu_dereference_protected(rt->fib6_node, |
| 2084 |
lockdep_is_held(&table->tb6_lock)); |
| 2085 |
if (!fn) |
| 2086 |
return -ENOENT; |
| 2087 |
|
| 2088 |
WARN_ON(!(fn->fn_flags & RTN_RTINFO)); |
| 2089 |
|
| 2090 |
/* |
| 2091 |
* Walk the leaf entries looking for ourself |
| 2092 |
*/ |
| 2093 |
|
| 2094 |
for (rtp = &fn->leaf; *rtp; rtp = rtp_next) { |
| 2095 |
struct fib6_info *cur = rcu_dereference_protected(*rtp, |
| 2096 |
lockdep_is_held(&table->tb6_lock)); |
| 2097 |
if (rt == cur) { |
| 2098 |
if (fib6_requires_src(cur)) |
| 2099 |
fib6_routes_require_src_dec(info->nl_net); |
| 2100 |
fib6_del_route(table, fn, rtp, info); |
| 2101 |
return 0; |
| 2102 |
} |
| 2103 |
rtp_next = &cur->fib6_next; |
| 2104 |
} |
| 2105 |
return -ENOENT; |
| 2106 |
} |
| 2107 |
|
| 2108 |
/* |
| 2109 |
* Tree traversal function. |
| 2110 |
* |
| 2111 |
* Certainly, it is not interrupt safe. |
| 2112 |
* However, it is internally reenterable wrt itself and fib6_add/fib6_del. |
| 2113 |
* It means, that we can modify tree during walking |
| 2114 |
* and use this function for garbage collection, clone pruning, |
| 2115 |
* cleaning tree when a device goes down etc. etc. |
| 2116 |
* |
| 2117 |
* It guarantees that every node will be traversed, |
| 2118 |
* and that it will be traversed only once. |
| 2119 |
* |
| 2120 |
* Callback function w->func may return: |
| 2121 |
* 0 -> continue walking. |
| 2122 |
* positive value -> walking is suspended (used by tree dumps, |
| 2123 |
* and probably by gc, if it will be split to several slices) |
| 2124 |
* negative value -> terminate walking. |
| 2125 |
* |
| 2126 |
* The function itself returns: |
| 2127 |
* 0 -> walk is complete. |
| 2128 |
* >0 -> walk is incomplete (i.e. suspended) |
| 2129 |
* <0 -> walk is terminated by an error. |
| 2130 |
* |
| 2131 |
* This function is called with tb6_lock held. |
| 2132 |
*/ |
| 2133 |
|
| 2134 |
static int fib6_walk_continue(struct fib6_walker *w) |
| 2135 |
{ |
| 2136 |
struct fib6_node *fn, *pn, *left, *right; |
| 2137 |
|
| 2138 |
/* w->root should always be table->tb6_root */ |
| 2139 |
WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT)); |
| 2140 |
|
| 2141 |
for (;;) { |
| 2142 |
fn = w->node; |
| 2143 |
if (!fn) |
| 2144 |
return 0; |
| 2145 |
|
| 2146 |
switch (w->state) { |
| 2147 |
#ifdef CONFIG_IPV6_SUBTREES |
| 2148 |
case FWS_S: |
| 2149 |
if (FIB6_SUBTREE(fn)) { |
| 2150 |
w->node = FIB6_SUBTREE(fn); |
| 2151 |
continue; |
| 2152 |
} |
| 2153 |
w->state = FWS_L; |
| 2154 |
fallthrough; |
| 2155 |
#endif |
| 2156 |
case FWS_L: |
| 2157 |
left = rcu_dereference_protected(fn->left, 1); |
| 2158 |
if (left) { |
| 2159 |
w->node = left; |
| 2160 |
w->state = FWS_INIT; |
| 2161 |
continue; |
| 2162 |
} |
| 2163 |
w->state = FWS_R; |
| 2164 |
fallthrough; |
| 2165 |
case FWS_R: |
| 2166 |
right = rcu_dereference_protected(fn->right, 1); |
| 2167 |
if (right) { |
| 2168 |
w->node = right; |
| 2169 |
w->state = FWS_INIT; |
| 2170 |
continue; |
| 2171 |
} |
| 2172 |
w->state = FWS_C; |
| 2173 |
w->leaf = rcu_dereference_protected(fn->leaf, 1); |
| 2174 |
fallthrough; |
| 2175 |
case FWS_C: |
| 2176 |
if (w->leaf && fn->fn_flags & RTN_RTINFO) { |
| 2177 |
int err; |
| 2178 |
|
| 2179 |
if (w->skip) { |
| 2180 |
w->skip--; |
| 2181 |
goto skip; |
| 2182 |
} |
| 2183 |
|
| 2184 |
err = w->func(w); |
| 2185 |
if (err) |
| 2186 |
return err; |
| 2187 |
|
| 2188 |
w->count++; |
| 2189 |
continue; |
| 2190 |
} |
| 2191 |
skip: |
| 2192 |
w->state = FWS_U; |
| 2193 |
fallthrough; |
| 2194 |
case FWS_U: |
| 2195 |
if (fn == w->root) |
| 2196 |
return 0; |
| 2197 |
pn = rcu_dereference_protected(fn->parent, 1); |
| 2198 |
left = rcu_dereference_protected(pn->left, 1); |
| 2199 |
right = rcu_dereference_protected(pn->right, 1); |
| 2200 |
w->node = pn; |
| 2201 |
#ifdef CONFIG_IPV6_SUBTREES |
| 2202 |
if (FIB6_SUBTREE(pn) == fn) { |
| 2203 |
WARN_ON(!(fn->fn_flags & RTN_ROOT)); |
| 2204 |
w->state = FWS_L; |
| 2205 |
continue; |
| 2206 |
} |
| 2207 |
#endif |
| 2208 |
if (left == fn) { |
| 2209 |
w->state = FWS_R; |
| 2210 |
continue; |
| 2211 |
} |
| 2212 |
if (right == fn) { |
| 2213 |
w->state = FWS_C; |
| 2214 |
w->leaf = rcu_dereference_protected(w->node->leaf, 1); |
| 2215 |
continue; |
| 2216 |
} |
| 2217 |
#if RT6_DEBUG >= 2 |
| 2218 |
WARN_ON(1); |
| 2219 |
#endif |
| 2220 |
} |
| 2221 |
} |
| 2222 |
} |
| 2223 |
|
| 2224 |
static int fib6_walk(struct net *net, struct fib6_walker *w) |
| 2225 |
{ |
| 2226 |
int res; |
| 2227 |
|
| 2228 |
w->state = FWS_INIT; |
| 2229 |
w->node = w->root; |
| 2230 |
|
| 2231 |
fib6_walker_link(net, w); |
| 2232 |
res = fib6_walk_continue(w); |
| 2233 |
if (res <= 0) |
| 2234 |
fib6_walker_unlink(net, w); |
| 2235 |
return res; |
| 2236 |
} |
| 2237 |
|
| 2238 |
static int fib6_clean_node(struct fib6_walker *w) |
| 2239 |
{ |
| 2240 |
int res; |
| 2241 |
struct fib6_info *rt; |
| 2242 |
struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w); |
| 2243 |
struct nl_info info = { |
| 2244 |
.nl_net = c->net, |
| 2245 |
.skip_notify = c->skip_notify, |
| 2246 |
}; |
| 2247 |
|
| 2248 |
if (c->sernum != FIB6_NO_SERNUM_CHANGE && |
| 2249 |
READ_ONCE(w->node->fn_sernum) != c->sernum) |
| 2250 |
WRITE_ONCE(w->node->fn_sernum, c->sernum); |
| 2251 |
|
| 2252 |
if (!c->func) { |
| 2253 |
WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE); |
| 2254 |
w->leaf = NULL; |
| 2255 |
return 0; |
| 2256 |
} |
| 2257 |
|
| 2258 |
for_each_fib6_walker_rt(w) { |
| 2259 |
res = c->func(rt, c->arg); |
| 2260 |
if (res == -1) { |
| 2261 |
w->leaf = rt; |
| 2262 |
res = fib6_del(rt, &info); |
| 2263 |
if (res) { |
| 2264 |
#if RT6_DEBUG >= 2 |
| 2265 |
pr_debug("%s: del failed: rt=%p@%p err=%d\n", |
| 2266 |
__func__, rt, |
| 2267 |
rcu_access_pointer(rt->fib6_node), |
| 2268 |
res); |
| 2269 |
#endif |
| 2270 |
continue; |
| 2271 |
} |
| 2272 |
return 0; |
| 2273 |
} else if (res == -2) { |
| 2274 |
if (WARN_ON(!rt->fib6_nsiblings)) |
| 2275 |
continue; |
| 2276 |
rt = list_last_entry(&rt->fib6_siblings, |
| 2277 |
struct fib6_info, fib6_siblings); |
| 2278 |
continue; |
| 2279 |
} |
| 2280 |
WARN_ON(res != 0); |
| 2281 |
} |
| 2282 |
w->leaf = rt; |
| 2283 |
return 0; |
| 2284 |
} |
| 2285 |
|
| 2286 |
/* |
| 2287 |
* Convenient frontend to tree walker. |
| 2288 |
* |
| 2289 |
* func is called on each route. |
| 2290 |
* It may return -2 -> skip multipath route. |
| 2291 |
* -1 -> delete this route. |
| 2292 |
* 0 -> continue walking |
| 2293 |
*/ |
| 2294 |
|
| 2295 |
static void fib6_clean_tree(struct net *net, struct fib6_node *root, |
| 2296 |
int (*func)(struct fib6_info *, void *arg), |
| 2297 |
int sernum, void *arg, bool skip_notify) |
| 2298 |
{ |
| 2299 |
struct fib6_cleaner c; |
| 2300 |
|
| 2301 |
c.w.root = root; |
| 2302 |
c.w.func = fib6_clean_node; |
| 2303 |
c.w.count = 0; |
| 2304 |
c.w.skip = 0; |
| 2305 |
c.w.skip_in_node = 0; |
| 2306 |
c.func = func; |
| 2307 |
c.sernum = sernum; |
| 2308 |
c.arg = arg; |
| 2309 |
c.net = net; |
| 2310 |
c.skip_notify = skip_notify; |
| 2311 |
|
| 2312 |
fib6_walk(net, &c.w); |
| 2313 |
} |
| 2314 |
|
| 2315 |
static void __fib6_clean_all(struct net *net, |
| 2316 |
int (*func)(struct fib6_info *, void *), |
| 2317 |
int sernum, void *arg, bool skip_notify) |
| 2318 |
{ |
| 2319 |
struct fib6_table *table; |
| 2320 |
struct hlist_head *head; |
| 2321 |
unsigned int h; |
| 2322 |
|
| 2323 |
rcu_read_lock(); |
| 2324 |
for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { |
| 2325 |
head = &net->ipv6.fib_table_hash[h]; |
| 2326 |
hlist_for_each_entry_rcu(table, head, tb6_hlist) { |
| 2327 |
spin_lock_bh(&table->tb6_lock); |
| 2328 |
fib6_clean_tree(net, &table->tb6_root, |
| 2329 |
func, sernum, arg, skip_notify); |
| 2330 |
spin_unlock_bh(&table->tb6_lock); |
| 2331 |
} |
| 2332 |
} |
| 2333 |
rcu_read_unlock(); |
| 2334 |
} |
| 2335 |
|
| 2336 |
void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *), |
| 2337 |
void *arg) |
| 2338 |
{ |
| 2339 |
__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false); |
| 2340 |
} |
| 2341 |
|
| 2342 |
void fib6_clean_all_skip_notify(struct net *net, |
| 2343 |
int (*func)(struct fib6_info *, void *), |
| 2344 |
void *arg) |
| 2345 |
{ |
| 2346 |
__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true); |
| 2347 |
} |
| 2348 |
|
| 2349 |
static void fib6_flush_trees(struct net *net) |
| 2350 |
{ |
| 2351 |
int new_sernum = fib6_new_sernum(net); |
| 2352 |
|
| 2353 |
__fib6_clean_all(net, NULL, new_sernum, NULL, false); |
| 2354 |
} |
| 2355 |
|
| 2356 |
/* |
| 2357 |
* Garbage collection |
| 2358 |
*/ |
| 2359 |
void fib6_age_exceptions(struct fib6_info *rt, struct fib6_gc_args *gc_args, |
| 2360 |
unsigned long now) |
| 2361 |
{ |
| 2362 |
bool may_expire = rt->fib6_flags & RTF_EXPIRES && rt->expires; |
| 2363 |
int old_more = gc_args->more; |
| 2364 |
|
| 2365 |
rt6_age_exceptions(rt, gc_args, now); |
| 2366 |
|
| 2367 |
if (!may_expire && old_more == gc_args->more) |
| 2368 |
fib6_remove_gc_list(rt); |
| 2369 |
} |
| 2370 |
|
| 2371 |
static int fib6_age(struct fib6_info *rt, struct fib6_gc_args *gc_args) |
| 2372 |
{ |
| 2373 |
unsigned long now = jiffies; |
| 2374 |
|
| 2375 |
/* |
| 2376 |
* check addrconf expiration here. |
| 2377 |
* Routes are expired even if they are in use. |
| 2378 |
*/ |
| 2379 |
|
| 2380 |
if (rt->fib6_flags & RTF_EXPIRES && rt->expires) { |
| 2381 |
if (time_after(now, rt->expires)) { |
| 2382 |
pr_debug("expiring %p\n", rt); |
| 2383 |
return -1; |
| 2384 |
} |
| 2385 |
gc_args->more++; |
| 2386 |
} |
| 2387 |
|
| 2388 |
/* Also age clones in the exception table. |
| 2389 |
* Note, that clones are aged out |
| 2390 |
* only if they are not in use now. |
| 2391 |
*/ |
| 2392 |
fib6_age_exceptions(rt, gc_args, now); |
| 2393 |
|
| 2394 |
return 0; |
| 2395 |
} |
| 2396 |
|
| 2397 |
static void fib6_gc_table(struct net *net, |
| 2398 |
struct fib6_table *tb6, |
| 2399 |
struct fib6_gc_args *gc_args) |
| 2400 |
{ |
| 2401 |
struct fib6_info *rt; |
| 2402 |
struct hlist_node *n; |
| 2403 |
struct nl_info info = { |
| 2404 |
.nl_net = net, |
| 2405 |
.skip_notify = false, |
| 2406 |
}; |
| 2407 |
|
| 2408 |
hlist_for_each_entry_safe(rt, n, &tb6->tb6_gc_hlist, gc_link) |
| 2409 |
if (fib6_age(rt, gc_args) == -1) |
| 2410 |
fib6_del(rt, &info); |
| 2411 |
} |
| 2412 |
|
| 2413 |
static void fib6_gc_all(struct net *net, struct fib6_gc_args *gc_args) |
| 2414 |
{ |
| 2415 |
struct fib6_table *table; |
| 2416 |
struct hlist_head *head; |
| 2417 |
unsigned int h; |
| 2418 |
|
| 2419 |
rcu_read_lock(); |
| 2420 |
for (h = 0; h < FIB6_TABLE_HASHSZ; h++) { |
| 2421 |
head = &net->ipv6.fib_table_hash[h]; |
| 2422 |
hlist_for_each_entry_rcu(table, head, tb6_hlist) { |
| 2423 |
spin_lock_bh(&table->tb6_lock); |
| 2424 |
|
| 2425 |
fib6_gc_table(net, table, gc_args); |
| 2426 |
|
| 2427 |
spin_unlock_bh(&table->tb6_lock); |
| 2428 |
} |
| 2429 |
} |
| 2430 |
rcu_read_unlock(); |
| 2431 |
} |
| 2432 |
|
| 2433 |
void fib6_run_gc(unsigned long expires, struct net *net, bool force) |
| 2434 |
{ |
| 2435 |
struct fib6_gc_args gc_args; |
| 2436 |
int ip6_rt_gc_interval; |
| 2437 |
unsigned long now; |
| 2438 |
|
| 2439 |
if (force) { |
| 2440 |
spin_lock_bh(&net->ipv6.fib6_gc_lock); |
| 2441 |
} else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) { |
| 2442 |
mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ); |
| 2443 |
return; |
| 2444 |
} |
| 2445 |
ip6_rt_gc_interval = READ_ONCE(net->ipv6.sysctl.ip6_rt_gc_interval); |
| 2446 |
gc_args.timeout = expires ? (int)expires : ip6_rt_gc_interval; |
| 2447 |
gc_args.more = 0; |
| 2448 |
|
| 2449 |
fib6_gc_all(net, &gc_args); |
| 2450 |
now = jiffies; |
| 2451 |
net->ipv6.ip6_rt_last_gc = now; |
| 2452 |
|
| 2453 |
if (gc_args.more) |
| 2454 |
mod_timer(&net->ipv6.ip6_fib_timer, |
| 2455 |
round_jiffies(now + ip6_rt_gc_interval)); |
| 2456 |
else |
| 2457 |
timer_delete(&net->ipv6.ip6_fib_timer); |
| 2458 |
spin_unlock_bh(&net->ipv6.fib6_gc_lock); |
| 2459 |
} |
| 2460 |
|
| 2461 |
static void fib6_gc_timer_cb(struct timer_list *t) |
| 2462 |
{ |
| 2463 |
struct net *arg = timer_container_of(arg, t, ipv6.ip6_fib_timer); |
| 2464 |
|
| 2465 |
fib6_run_gc(0, arg, true); |
| 2466 |
} |
| 2467 |
|
| 2468 |
static int __net_init fib6_net_init(struct net *net) |
| 2469 |
{ |
| 2470 |
size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ; |
| 2471 |
int err; |
| 2472 |
|
| 2473 |
err = fib6_notifier_init(net); |
| 2474 |
if (err) |
| 2475 |
return err; |
| 2476 |
|
| 2477 |
/* Default to 3-tuple */ |
| 2478 |
net->ipv6.sysctl.multipath_hash_fields = |
| 2479 |
FIB_MULTIPATH_HASH_FIELD_DEFAULT_MASK; |
| 2480 |
|
| 2481 |
spin_lock_init(&net->ipv6.fib6_gc_lock); |
| 2482 |
rwlock_init(&net->ipv6.fib6_walker_lock); |
| 2483 |
INIT_LIST_HEAD(&net->ipv6.fib6_walkers); |
| 2484 |
timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0); |
| 2485 |
|
| 2486 |
net->ipv6.rt6_stats = kzalloc_obj(*net->ipv6.rt6_stats); |
| 2487 |
if (!net->ipv6.rt6_stats) |
| 2488 |
goto out_notifier; |
| 2489 |
|
| 2490 |
/* Avoid false sharing : Use at least a full cache line */ |
| 2491 |
size = max_t(size_t, size, L1_CACHE_BYTES); |
| 2492 |
|
| 2493 |
net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL); |
| 2494 |
if (!net->ipv6.fib_table_hash) |
| 2495 |
goto out_rt6_stats; |
| 2496 |
|
| 2497 |
spin_lock_init(&net->ipv6.fib_table_hash_lock); |
| 2498 |
|
| 2499 |
net->ipv6.fib6_main_tbl = kzalloc_obj(*net->ipv6.fib6_main_tbl); |
| 2500 |
if (!net->ipv6.fib6_main_tbl) |
| 2501 |
goto out_fib_table_hash; |
| 2502 |
|
| 2503 |
net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN; |
| 2504 |
rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf, |
| 2505 |
net->ipv6.fib6_null_entry); |
| 2506 |
net->ipv6.fib6_main_tbl->tb6_root.fn_flags = |
| 2507 |
RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; |
| 2508 |
inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers); |
| 2509 |
INIT_HLIST_HEAD(&net->ipv6.fib6_main_tbl->tb6_gc_hlist); |
| 2510 |
|
| 2511 |
#ifdef CONFIG_IPV6_MULTIPLE_TABLES |
| 2512 |
net->ipv6.fib6_local_tbl = kzalloc_obj(*net->ipv6.fib6_local_tbl); |
| 2513 |
if (!net->ipv6.fib6_local_tbl) |
| 2514 |
goto out_fib6_main_tbl; |
| 2515 |
net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL; |
| 2516 |
rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf, |
| 2517 |
net->ipv6.fib6_null_entry); |
| 2518 |
net->ipv6.fib6_local_tbl->tb6_root.fn_flags = |
| 2519 |
RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO; |
| 2520 |
inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers); |
| 2521 |
INIT_HLIST_HEAD(&net->ipv6.fib6_local_tbl->tb6_gc_hlist); |
| 2522 |
#endif |
| 2523 |
fib6_tables_init(net); |
| 2524 |
|
| 2525 |
return 0; |
| 2526 |
|
| 2527 |
#ifdef CONFIG_IPV6_MULTIPLE_TABLES |
| 2528 |
out_fib6_main_tbl: |
| 2529 |
kfree(net->ipv6.fib6_main_tbl); |
| 2530 |
#endif |
| 2531 |
out_fib_table_hash: |
| 2532 |
kfree(net->ipv6.fib_table_hash); |
| 2533 |
out_rt6_stats: |
| 2534 |
kfree(net->ipv6.rt6_stats); |
| 2535 |
out_notifier: |
| 2536 |
fib6_notifier_exit(net); |
| 2537 |
return -ENOMEM; |
| 2538 |
} |
| 2539 |
|
| 2540 |
static void fib6_net_exit(struct net *net) |
| 2541 |
{ |
| 2542 |
unsigned int i; |
| 2543 |
|
| 2544 |
timer_delete_sync(&net->ipv6.ip6_fib_timer); |
| 2545 |
|
| 2546 |
for (i = 0; i < FIB6_TABLE_HASHSZ; i++) { |
| 2547 |
struct hlist_head *head = &net->ipv6.fib_table_hash[i]; |
| 2548 |
struct hlist_node *tmp; |
| 2549 |
struct fib6_table *tb; |
| 2550 |
|
| 2551 |
hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) { |
| 2552 |
hlist_del(&tb->tb6_hlist); |
| 2553 |
fib6_free_table(tb); |
| 2554 |
} |
| 2555 |
} |
| 2556 |
|
| 2557 |
kfree(net->ipv6.fib_table_hash); |
| 2558 |
kfree(net->ipv6.rt6_stats); |
| 2559 |
fib6_notifier_exit(net); |
| 2560 |
} |
| 2561 |
|
| 2562 |
static struct pernet_operations fib6_net_ops = { |
| 2563 |
.init = fib6_net_init, |
| 2564 |
.exit = fib6_net_exit, |
| 2565 |
}; |
| 2566 |
|
| 2567 |
static const struct rtnl_msg_handler fib6_rtnl_msg_handlers[] __initconst_or_module = { |
| 2568 |
{.owner = THIS_MODULE, .protocol = PF_INET6, .msgtype = RTM_GETROUTE, |
| 2569 |
.dumpit = inet6_dump_fib, |
| 2570 |
.flags = RTNL_FLAG_DUMP_UNLOCKED | RTNL_FLAG_DUMP_SPLIT_NLM_DONE}, |
| 2571 |
}; |
| 2572 |
|
| 2573 |
int __init fib6_init(void) |
| 2574 |
{ |
| 2575 |
int ret = -ENOMEM; |
| 2576 |
|
| 2577 |
fib6_node_kmem = KMEM_CACHE(fib6_node, |
| 2578 |
SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT); |
| 2579 |
if (!fib6_node_kmem) |
| 2580 |
goto out; |
| 2581 |
|
| 2582 |
ret = register_pernet_subsys(&fib6_net_ops); |
| 2583 |
if (ret) |
| 2584 |
goto out_kmem_cache_create; |
| 2585 |
|
| 2586 |
ret = rtnl_register_many(fib6_rtnl_msg_handlers); |
| 2587 |
if (ret) |
| 2588 |
goto out_unregister_subsys; |
| 2589 |
|
| 2590 |
__fib6_flush_trees = fib6_flush_trees; |
| 2591 |
out: |
| 2592 |
return ret; |
| 2593 |
|
| 2594 |
out_unregister_subsys: |
| 2595 |
unregister_pernet_subsys(&fib6_net_ops); |
| 2596 |
out_kmem_cache_create: |
| 2597 |
kmem_cache_destroy(fib6_node_kmem); |
| 2598 |
goto out; |
| 2599 |
} |
| 2600 |
|
| 2601 |
void fib6_gc_cleanup(void) |
| 2602 |
{ |
| 2603 |
unregister_pernet_subsys(&fib6_net_ops); |
| 2604 |
kmem_cache_destroy(fib6_node_kmem); |
| 2605 |
} |
| 2606 |
|
| 2607 |
#ifdef CONFIG_PROC_FS |
| 2608 |
static int ipv6_route_native_seq_show(struct seq_file *seq, void *v) |
| 2609 |
{ |
| 2610 |
struct fib6_info *rt = v; |
| 2611 |
struct ipv6_route_iter *iter = seq->private; |
| 2612 |
struct fib6_nh *fib6_nh = rt->fib6_nh; |
| 2613 |
unsigned int flags = rt->fib6_flags; |
| 2614 |
const struct net_device *dev; |
| 2615 |
|
| 2616 |
if (rt->nh) |
| 2617 |
fib6_nh = nexthop_fib6_nh(rt->nh); |
| 2618 |
|
| 2619 |
seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen); |
| 2620 |
|
| 2621 |
#ifdef CONFIG_IPV6_SUBTREES |
| 2622 |
seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen); |
| 2623 |
#else |
| 2624 |
seq_puts(seq, "00000000000000000000000000000000 00 "); |
| 2625 |
#endif |
| 2626 |
if (fib6_nh->fib_nh_gw_family) { |
| 2627 |
flags |= RTF_GATEWAY; |
| 2628 |
seq_printf(seq, "%pi6", &fib6_nh->fib_nh_gw6); |
| 2629 |
} else { |
| 2630 |
seq_puts(seq, "00000000000000000000000000000000"); |
| 2631 |
} |
| 2632 |
|
| 2633 |
dev = fib6_nh->fib_nh_dev; |
| 2634 |
seq_printf(seq, " %08x %08x %08x %08x %8s\n", |
| 2635 |
rt->fib6_metric, refcount_read(&rt->fib6_ref), 0, |
| 2636 |
flags, dev ? dev->name : ""); |
| 2637 |
iter->w.leaf = NULL; |
| 2638 |
return 0; |
| 2639 |
} |
| 2640 |
|
| 2641 |
static int ipv6_route_yield(struct fib6_walker *w) |
| 2642 |
{ |
| 2643 |
struct ipv6_route_iter *iter = w->args; |
| 2644 |
|
| 2645 |
if (!iter->skip) |
| 2646 |
return 1; |
| 2647 |
|
| 2648 |
do { |
| 2649 |
iter->w.leaf = rcu_dereference_protected( |
| 2650 |
iter->w.leaf->fib6_next, |
| 2651 |
lockdep_is_held(&iter->tbl->tb6_lock)); |
| 2652 |
iter->skip--; |
| 2653 |
if (!iter->skip && iter->w.leaf) |
| 2654 |
return 1; |
| 2655 |
} while (iter->w.leaf); |
| 2656 |
|
| 2657 |
return 0; |
| 2658 |
} |
| 2659 |
|
| 2660 |
static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter, |
| 2661 |
struct net *net) |
| 2662 |
{ |
| 2663 |
memset(&iter->w, 0, sizeof(iter->w)); |
| 2664 |
iter->w.func = ipv6_route_yield; |
| 2665 |
iter->w.root = &iter->tbl->tb6_root; |
| 2666 |
iter->w.state = FWS_INIT; |
| 2667 |
iter->w.node = iter->w.root; |
| 2668 |
iter->w.args = iter; |
| 2669 |
iter->sernum = READ_ONCE(iter->w.root->fn_sernum); |
| 2670 |
INIT_LIST_HEAD(&iter->w.lh); |
| 2671 |
fib6_walker_link(net, &iter->w); |
| 2672 |
} |
| 2673 |
|
| 2674 |
static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl, |
| 2675 |
struct net *net) |
| 2676 |
{ |
| 2677 |
unsigned int h; |
| 2678 |
struct hlist_node *node; |
| 2679 |
|
| 2680 |
if (tbl) { |
| 2681 |
h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1; |
| 2682 |
node = rcu_dereference(hlist_next_rcu(&tbl->tb6_hlist)); |
| 2683 |
} else { |
| 2684 |
h = 0; |
| 2685 |
node = NULL; |
| 2686 |
} |
| 2687 |
|
| 2688 |
while (!node && h < FIB6_TABLE_HASHSZ) { |
| 2689 |
node = rcu_dereference( |
| 2690 |
hlist_first_rcu(&net->ipv6.fib_table_hash[h++])); |
| 2691 |
} |
| 2692 |
return hlist_entry_safe(node, struct fib6_table, tb6_hlist); |
| 2693 |
} |
| 2694 |
|
| 2695 |
static void ipv6_route_check_sernum(struct ipv6_route_iter *iter) |
| 2696 |
{ |
| 2697 |
int sernum = READ_ONCE(iter->w.root->fn_sernum); |
| 2698 |
|
| 2699 |
if (iter->sernum != sernum) { |
| 2700 |
iter->sernum = sernum; |
| 2701 |
iter->w.state = FWS_INIT; |
| 2702 |
iter->w.node = iter->w.root; |
| 2703 |
WARN_ON(iter->w.skip); |
| 2704 |
iter->w.skip = iter->w.count; |
| 2705 |
} |
| 2706 |
} |
| 2707 |
|
| 2708 |
static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos) |
| 2709 |
{ |
| 2710 |
int r; |
| 2711 |
struct fib6_info *n; |
| 2712 |
struct net *net = seq_file_net(seq); |
| 2713 |
struct ipv6_route_iter *iter = seq->private; |
| 2714 |
|
| 2715 |
++(*pos); |
| 2716 |
if (!v) |
| 2717 |
goto iter_table; |
| 2718 |
|
| 2719 |
n = rcu_dereference(((struct fib6_info *)v)->fib6_next); |
| 2720 |
if (n) |
| 2721 |
return n; |
| 2722 |
|
| 2723 |
iter_table: |
| 2724 |
ipv6_route_check_sernum(iter); |
| 2725 |
spin_lock_bh(&iter->tbl->tb6_lock); |
| 2726 |
r = fib6_walk_continue(&iter->w); |
| 2727 |
spin_unlock_bh(&iter->tbl->tb6_lock); |
| 2728 |
if (r > 0) { |
| 2729 |
return iter->w.leaf; |
| 2730 |
} else if (r < 0) { |
| 2731 |
fib6_walker_unlink(net, &iter->w); |
| 2732 |
return NULL; |
| 2733 |
} |
| 2734 |
fib6_walker_unlink(net, &iter->w); |
| 2735 |
|
| 2736 |
iter->tbl = ipv6_route_seq_next_table(iter->tbl, net); |
| 2737 |
if (!iter->tbl) |
| 2738 |
return NULL; |
| 2739 |
|
| 2740 |
ipv6_route_seq_setup_walk(iter, net); |
| 2741 |
goto iter_table; |
| 2742 |
} |
| 2743 |
|
| 2744 |
static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos) |
| 2745 |
__acquires(RCU) |
| 2746 |
{ |
| 2747 |
struct net *net = seq_file_net(seq); |
| 2748 |
struct ipv6_route_iter *iter = seq->private; |
| 2749 |
|
| 2750 |
rcu_read_lock(); |
| 2751 |
iter->tbl = ipv6_route_seq_next_table(NULL, net); |
| 2752 |
iter->skip = *pos; |
| 2753 |
|
| 2754 |
if (iter->tbl) { |
| 2755 |
loff_t p = 0; |
| 2756 |
|
| 2757 |
ipv6_route_seq_setup_walk(iter, net); |
| 2758 |
return ipv6_route_seq_next(seq, NULL, &p); |
| 2759 |
} else { |
| 2760 |
return NULL; |
| 2761 |
} |
| 2762 |
} |
| 2763 |
|
| 2764 |
static bool ipv6_route_iter_active(struct ipv6_route_iter *iter) |
| 2765 |
{ |
| 2766 |
struct fib6_walker *w = &iter->w; |
| 2767 |
return w->node && !(w->state == FWS_U && w->node == w->root); |
| 2768 |
} |
| 2769 |
|
| 2770 |
static void ipv6_route_native_seq_stop(struct seq_file *seq, void *v) |
| 2771 |
__releases(RCU) |
| 2772 |
{ |
| 2773 |
struct net *net = seq_file_net(seq); |
| 2774 |
struct ipv6_route_iter *iter = seq->private; |
| 2775 |
|
| 2776 |
if (ipv6_route_iter_active(iter)) |
| 2777 |
fib6_walker_unlink(net, &iter->w); |
| 2778 |
|
| 2779 |
rcu_read_unlock(); |
| 2780 |
} |
| 2781 |
|
| 2782 |
#if IS_BUILTIN(CONFIG_IPV6) && defined(CONFIG_BPF_SYSCALL) |
| 2783 |
static int ipv6_route_prog_seq_show(struct bpf_prog *prog, |
| 2784 |
struct bpf_iter_meta *meta, |
| 2785 |
void *v) |
| 2786 |
{ |
| 2787 |
struct bpf_iter__ipv6_route ctx; |
| 2788 |
|
| 2789 |
ctx.meta = meta; |
| 2790 |
ctx.rt = v; |
| 2791 |
return bpf_iter_run_prog(prog, &ctx); |
| 2792 |
} |
| 2793 |
|
| 2794 |
static int ipv6_route_seq_show(struct seq_file *seq, void *v) |
| 2795 |
{ |
| 2796 |
struct ipv6_route_iter *iter = seq->private; |
| 2797 |
struct bpf_iter_meta meta; |
| 2798 |
struct bpf_prog *prog; |
| 2799 |
int ret; |
| 2800 |
|
| 2801 |
meta.seq = seq; |
| 2802 |
prog = bpf_iter_get_info(&meta, false); |
| 2803 |
if (!prog) |
| 2804 |
return ipv6_route_native_seq_show(seq, v); |
| 2805 |
|
| 2806 |
ret = ipv6_route_prog_seq_show(prog, &meta, v); |
| 2807 |
iter->w.leaf = NULL; |
| 2808 |
|
| 2809 |
return ret; |
| 2810 |
} |
| 2811 |
|
| 2812 |
static void ipv6_route_seq_stop(struct seq_file *seq, void *v) |
| 2813 |
{ |
| 2814 |
struct bpf_iter_meta meta; |
| 2815 |
struct bpf_prog *prog; |
| 2816 |
|
| 2817 |
if (!v) { |
| 2818 |
meta.seq = seq; |
| 2819 |
prog = bpf_iter_get_info(&meta, true); |
| 2820 |
if (prog) |
| 2821 |
(void)ipv6_route_prog_seq_show(prog, &meta, v); |
| 2822 |
} |
| 2823 |
|
| 2824 |
ipv6_route_native_seq_stop(seq, v); |
| 2825 |
} |
| 2826 |
#else |
| 2827 |
static int ipv6_route_seq_show(struct seq_file *seq, void *v) |
| 2828 |
{ |
| 2829 |
return ipv6_route_native_seq_show(seq, v); |
| 2830 |
} |
| 2831 |
|
| 2832 |
static void ipv6_route_seq_stop(struct seq_file *seq, void *v) |
| 2833 |
{ |
| 2834 |
ipv6_route_native_seq_stop(seq, v); |
| 2835 |
} |
| 2836 |
#endif |
| 2837 |
|
| 2838 |
const struct seq_operations ipv6_route_seq_ops = { |
| 2839 |
.start = ipv6_route_seq_start, |
| 2840 |
.next = ipv6_route_seq_next, |
| 2841 |
.stop = ipv6_route_seq_stop, |
| 2842 |
.show = ipv6_route_seq_show |
| 2843 |
}; |
| 2844 |
#endif /* CONFIG_PROC_FS */ |