Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Linux INET6 implementation |
| 3 | * Forwarding Information Database |
| 4 | * |
| 5 | * Authors: |
| 6 | * Pedro Roque <roque@di.fc.ul.pt> |
| 7 | * |
| 8 | * $Id: ip6_fib.c,v 1.25 2001/10/31 21:55:55 davem Exp $ |
| 9 | * |
| 10 | * This program is free software; you can redistribute it and/or |
| 11 | * modify it under the terms of the GNU General Public License |
| 12 | * as published by the Free Software Foundation; either version |
| 13 | * 2 of the License, or (at your option) any later version. |
| 14 | */ |
| 15 | |
| 16 | /* |
| 17 | * Changes: |
| 18 | * Yuji SEKIYA @USAGI: Support default route on router node; |
| 19 | * remove ip6_null_entry from the top of |
| 20 | * routing table. |
| 21 | */ |
| 22 | #include <linux/config.h> |
| 23 | #include <linux/errno.h> |
| 24 | #include <linux/types.h> |
| 25 | #include <linux/net.h> |
| 26 | #include <linux/route.h> |
| 27 | #include <linux/netdevice.h> |
| 28 | #include <linux/in6.h> |
| 29 | #include <linux/init.h> |
| 30 | |
| 31 | #ifdef CONFIG_PROC_FS |
| 32 | #include <linux/proc_fs.h> |
| 33 | #endif |
| 34 | |
| 35 | #include <net/ipv6.h> |
| 36 | #include <net/ndisc.h> |
| 37 | #include <net/addrconf.h> |
| 38 | |
| 39 | #include <net/ip6_fib.h> |
| 40 | #include <net/ip6_route.h> |
| 41 | |
| 42 | #define RT6_DEBUG 2 |
| 43 | |
| 44 | #if RT6_DEBUG >= 3 |
| 45 | #define RT6_TRACE(x...) printk(KERN_DEBUG x) |
| 46 | #else |
| 47 | #define RT6_TRACE(x...) do { ; } while (0) |
| 48 | #endif |
| 49 | |
| 50 | struct rt6_statistics rt6_stats; |
| 51 | |
| 52 | static kmem_cache_t * fib6_node_kmem; |
| 53 | |
| 54 | enum fib_walk_state_t |
| 55 | { |
| 56 | #ifdef CONFIG_IPV6_SUBTREES |
| 57 | FWS_S, |
| 58 | #endif |
| 59 | FWS_L, |
| 60 | FWS_R, |
| 61 | FWS_C, |
| 62 | FWS_U |
| 63 | }; |
| 64 | |
| 65 | struct fib6_cleaner_t |
| 66 | { |
| 67 | struct fib6_walker_t w; |
| 68 | int (*func)(struct rt6_info *, void *arg); |
| 69 | void *arg; |
| 70 | }; |
| 71 | |
| 72 | DEFINE_RWLOCK(fib6_walker_lock); |
| 73 | |
| 74 | |
| 75 | #ifdef CONFIG_IPV6_SUBTREES |
| 76 | #define FWS_INIT FWS_S |
| 77 | #define SUBTREE(fn) ((fn)->subtree) |
| 78 | #else |
| 79 | #define FWS_INIT FWS_L |
| 80 | #define SUBTREE(fn) NULL |
| 81 | #endif |
| 82 | |
| 83 | static void fib6_prune_clones(struct fib6_node *fn, struct rt6_info *rt); |
| 84 | static struct fib6_node * fib6_repair_tree(struct fib6_node *fn); |
| 85 | |
| 86 | /* |
| 87 | * A routing update causes an increase of the serial number on the |
| 88 | * affected subtree. This allows for cached routes to be asynchronously |
| 89 | * tested when modifications are made to the destination cache as a |
| 90 | * result of redirects, path MTU changes, etc. |
| 91 | */ |
| 92 | |
| 93 | static __u32 rt_sernum; |
| 94 | |
| 95 | static struct timer_list ip6_fib_timer = TIMER_INITIALIZER(fib6_run_gc, 0, 0); |
| 96 | |
| 97 | struct fib6_walker_t fib6_walker_list = { |
| 98 | .prev = &fib6_walker_list, |
| 99 | .next = &fib6_walker_list, |
| 100 | }; |
| 101 | |
| 102 | #define FOR_WALKERS(w) for ((w)=fib6_walker_list.next; (w) != &fib6_walker_list; (w)=(w)->next) |
| 103 | |
| 104 | static __inline__ u32 fib6_new_sernum(void) |
| 105 | { |
| 106 | u32 n = ++rt_sernum; |
| 107 | if ((__s32)n <= 0) |
| 108 | rt_sernum = n = 1; |
| 109 | return n; |
| 110 | } |
| 111 | |
| 112 | /* |
| 113 | * Auxiliary address test functions for the radix tree. |
| 114 | * |
| 115 | * These assume a 32bit processor (although it will work on |
| 116 | * 64bit processors) |
| 117 | */ |
| 118 | |
| 119 | /* |
| 120 | * test bit |
| 121 | */ |
| 122 | |
| 123 | static __inline__ int addr_bit_set(void *token, int fn_bit) |
| 124 | { |
| 125 | __u32 *addr = token; |
| 126 | |
| 127 | return htonl(1 << ((~fn_bit)&0x1F)) & addr[fn_bit>>5]; |
| 128 | } |
| 129 | |
| 130 | /* |
| 131 | * find the first different bit between two addresses |
| 132 | * length of address must be a multiple of 32bits |
| 133 | */ |
| 134 | |
| 135 | static __inline__ int addr_diff(void *token1, void *token2, int addrlen) |
| 136 | { |
| 137 | __u32 *a1 = token1; |
| 138 | __u32 *a2 = token2; |
| 139 | int i; |
| 140 | |
| 141 | addrlen >>= 2; |
| 142 | |
| 143 | for (i = 0; i < addrlen; i++) { |
| 144 | __u32 xb; |
| 145 | |
| 146 | xb = a1[i] ^ a2[i]; |
| 147 | |
| 148 | if (xb) { |
| 149 | int j = 31; |
| 150 | |
| 151 | xb = ntohl(xb); |
| 152 | |
| 153 | while ((xb & (1 << j)) == 0) |
| 154 | j--; |
| 155 | |
| 156 | return (i * 32 + 31 - j); |
| 157 | } |
| 158 | } |
| 159 | |
| 160 | /* |
| 161 | * we should *never* get to this point since that |
| 162 | * would mean the addrs are equal |
| 163 | * |
| 164 | * However, we do get to it 8) And exacly, when |
| 165 | * addresses are equal 8) |
| 166 | * |
| 167 | * ip route add 1111::/128 via ... |
| 168 | * ip route add 1111::/64 via ... |
| 169 | * and we are here. |
| 170 | * |
| 171 | * Ideally, this function should stop comparison |
| 172 | * at prefix length. It does not, but it is still OK, |
| 173 | * if returned value is greater than prefix length. |
| 174 | * --ANK (980803) |
| 175 | */ |
| 176 | |
| 177 | return addrlen<<5; |
| 178 | } |
| 179 | |
| 180 | static __inline__ struct fib6_node * node_alloc(void) |
| 181 | { |
| 182 | struct fib6_node *fn; |
| 183 | |
| 184 | if ((fn = kmem_cache_alloc(fib6_node_kmem, SLAB_ATOMIC)) != NULL) |
| 185 | memset(fn, 0, sizeof(struct fib6_node)); |
| 186 | |
| 187 | return fn; |
| 188 | } |
| 189 | |
| 190 | static __inline__ void node_free(struct fib6_node * fn) |
| 191 | { |
| 192 | kmem_cache_free(fib6_node_kmem, fn); |
| 193 | } |
| 194 | |
| 195 | static __inline__ void rt6_release(struct rt6_info *rt) |
| 196 | { |
| 197 | if (atomic_dec_and_test(&rt->rt6i_ref)) |
| 198 | dst_free(&rt->u.dst); |
| 199 | } |
| 200 | |
| 201 | |
| 202 | /* |
| 203 | * Routing Table |
| 204 | * |
| 205 | * return the appropriate node for a routing tree "add" operation |
| 206 | * by either creating and inserting or by returning an existing |
| 207 | * node. |
| 208 | */ |
| 209 | |
| 210 | static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr, |
| 211 | int addrlen, int plen, |
| 212 | int offset) |
| 213 | { |
| 214 | struct fib6_node *fn, *in, *ln; |
| 215 | struct fib6_node *pn = NULL; |
| 216 | struct rt6key *key; |
| 217 | int bit; |
| 218 | int dir = 0; |
| 219 | __u32 sernum = fib6_new_sernum(); |
| 220 | |
| 221 | RT6_TRACE("fib6_add_1\n"); |
| 222 | |
| 223 | /* insert node in tree */ |
| 224 | |
| 225 | fn = root; |
| 226 | |
| 227 | do { |
| 228 | key = (struct rt6key *)((u8 *)fn->leaf + offset); |
| 229 | |
| 230 | /* |
| 231 | * Prefix match |
| 232 | */ |
| 233 | if (plen < fn->fn_bit || |
| 234 | !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) |
| 235 | goto insert_above; |
| 236 | |
| 237 | /* |
| 238 | * Exact match ? |
| 239 | */ |
| 240 | |
| 241 | if (plen == fn->fn_bit) { |
| 242 | /* clean up an intermediate node */ |
| 243 | if ((fn->fn_flags & RTN_RTINFO) == 0) { |
| 244 | rt6_release(fn->leaf); |
| 245 | fn->leaf = NULL; |
| 246 | } |
| 247 | |
| 248 | fn->fn_sernum = sernum; |
| 249 | |
| 250 | return fn; |
| 251 | } |
| 252 | |
| 253 | /* |
| 254 | * We have more bits to go |
| 255 | */ |
| 256 | |
| 257 | /* Try to walk down on tree. */ |
| 258 | fn->fn_sernum = sernum; |
| 259 | dir = addr_bit_set(addr, fn->fn_bit); |
| 260 | pn = fn; |
| 261 | fn = dir ? fn->right: fn->left; |
| 262 | } while (fn); |
| 263 | |
| 264 | /* |
| 265 | * We walked to the bottom of tree. |
| 266 | * Create new leaf node without children. |
| 267 | */ |
| 268 | |
| 269 | ln = node_alloc(); |
| 270 | |
| 271 | if (ln == NULL) |
| 272 | return NULL; |
| 273 | ln->fn_bit = plen; |
| 274 | |
| 275 | ln->parent = pn; |
| 276 | ln->fn_sernum = sernum; |
| 277 | |
| 278 | if (dir) |
| 279 | pn->right = ln; |
| 280 | else |
| 281 | pn->left = ln; |
| 282 | |
| 283 | return ln; |
| 284 | |
| 285 | |
| 286 | insert_above: |
| 287 | /* |
| 288 | * split since we don't have a common prefix anymore or |
| 289 | * we have a less significant route. |
| 290 | * we've to insert an intermediate node on the list |
| 291 | * this new node will point to the one we need to create |
| 292 | * and the current |
| 293 | */ |
| 294 | |
| 295 | pn = fn->parent; |
| 296 | |
| 297 | /* find 1st bit in difference between the 2 addrs. |
| 298 | |
| 299 | See comment in addr_diff: bit may be an invalid value, |
| 300 | but if it is >= plen, the value is ignored in any case. |
| 301 | */ |
| 302 | |
| 303 | bit = addr_diff(addr, &key->addr, addrlen); |
| 304 | |
| 305 | /* |
| 306 | * (intermediate)[in] |
| 307 | * / \ |
| 308 | * (new leaf node)[ln] (old node)[fn] |
| 309 | */ |
| 310 | if (plen > bit) { |
| 311 | in = node_alloc(); |
| 312 | ln = node_alloc(); |
| 313 | |
| 314 | if (in == NULL || ln == NULL) { |
| 315 | if (in) |
| 316 | node_free(in); |
| 317 | if (ln) |
| 318 | node_free(ln); |
| 319 | return NULL; |
| 320 | } |
| 321 | |
| 322 | /* |
| 323 | * new intermediate node. |
| 324 | * RTN_RTINFO will |
| 325 | * be off since that an address that chooses one of |
| 326 | * the branches would not match less specific routes |
| 327 | * in the other branch |
| 328 | */ |
| 329 | |
| 330 | in->fn_bit = bit; |
| 331 | |
| 332 | in->parent = pn; |
| 333 | in->leaf = fn->leaf; |
| 334 | atomic_inc(&in->leaf->rt6i_ref); |
| 335 | |
| 336 | in->fn_sernum = sernum; |
| 337 | |
| 338 | /* update parent pointer */ |
| 339 | if (dir) |
| 340 | pn->right = in; |
| 341 | else |
| 342 | pn->left = in; |
| 343 | |
| 344 | ln->fn_bit = plen; |
| 345 | |
| 346 | ln->parent = in; |
| 347 | fn->parent = in; |
| 348 | |
| 349 | ln->fn_sernum = sernum; |
| 350 | |
| 351 | if (addr_bit_set(addr, bit)) { |
| 352 | in->right = ln; |
| 353 | in->left = fn; |
| 354 | } else { |
| 355 | in->left = ln; |
| 356 | in->right = fn; |
| 357 | } |
| 358 | } else { /* plen <= bit */ |
| 359 | |
| 360 | /* |
| 361 | * (new leaf node)[ln] |
| 362 | * / \ |
| 363 | * (old node)[fn] NULL |
| 364 | */ |
| 365 | |
| 366 | ln = node_alloc(); |
| 367 | |
| 368 | if (ln == NULL) |
| 369 | return NULL; |
| 370 | |
| 371 | ln->fn_bit = plen; |
| 372 | |
| 373 | ln->parent = pn; |
| 374 | |
| 375 | ln->fn_sernum = sernum; |
| 376 | |
| 377 | if (dir) |
| 378 | pn->right = ln; |
| 379 | else |
| 380 | pn->left = ln; |
| 381 | |
| 382 | if (addr_bit_set(&key->addr, plen)) |
| 383 | ln->right = fn; |
| 384 | else |
| 385 | ln->left = fn; |
| 386 | |
| 387 | fn->parent = ln; |
| 388 | } |
| 389 | return ln; |
| 390 | } |
| 391 | |
| 392 | /* |
| 393 | * Insert routing information in a node. |
| 394 | */ |
| 395 | |
| 396 | static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt, |
| 397 | struct nlmsghdr *nlh) |
| 398 | { |
| 399 | struct rt6_info *iter = NULL; |
| 400 | struct rt6_info **ins; |
| 401 | |
| 402 | ins = &fn->leaf; |
| 403 | |
| 404 | if (fn->fn_flags&RTN_TL_ROOT && |
| 405 | fn->leaf == &ip6_null_entry && |
| 406 | !(rt->rt6i_flags & (RTF_DEFAULT | RTF_ADDRCONF)) ){ |
| 407 | fn->leaf = rt; |
| 408 | rt->u.next = NULL; |
| 409 | goto out; |
| 410 | } |
| 411 | |
| 412 | for (iter = fn->leaf; iter; iter=iter->u.next) { |
| 413 | /* |
| 414 | * Search for duplicates |
| 415 | */ |
| 416 | |
| 417 | if (iter->rt6i_metric == rt->rt6i_metric) { |
| 418 | /* |
| 419 | * Same priority level |
| 420 | */ |
| 421 | |
| 422 | if (iter->rt6i_dev == rt->rt6i_dev && |
| 423 | iter->rt6i_idev == rt->rt6i_idev && |
| 424 | ipv6_addr_equal(&iter->rt6i_gateway, |
| 425 | &rt->rt6i_gateway)) { |
| 426 | if (!(iter->rt6i_flags&RTF_EXPIRES)) |
| 427 | return -EEXIST; |
| 428 | iter->rt6i_expires = rt->rt6i_expires; |
| 429 | if (!(rt->rt6i_flags&RTF_EXPIRES)) { |
| 430 | iter->rt6i_flags &= ~RTF_EXPIRES; |
| 431 | iter->rt6i_expires = 0; |
| 432 | } |
| 433 | return -EEXIST; |
| 434 | } |
| 435 | } |
| 436 | |
| 437 | if (iter->rt6i_metric > rt->rt6i_metric) |
| 438 | break; |
| 439 | |
| 440 | ins = &iter->u.next; |
| 441 | } |
| 442 | |
| 443 | /* |
| 444 | * insert node |
| 445 | */ |
| 446 | |
| 447 | out: |
| 448 | rt->u.next = iter; |
| 449 | *ins = rt; |
| 450 | rt->rt6i_node = fn; |
| 451 | atomic_inc(&rt->rt6i_ref); |
| 452 | inet6_rt_notify(RTM_NEWROUTE, rt, nlh); |
| 453 | rt6_stats.fib_rt_entries++; |
| 454 | |
| 455 | if ((fn->fn_flags & RTN_RTINFO) == 0) { |
| 456 | rt6_stats.fib_route_nodes++; |
| 457 | fn->fn_flags |= RTN_RTINFO; |
| 458 | } |
| 459 | |
| 460 | return 0; |
| 461 | } |
| 462 | |
| 463 | static __inline__ void fib6_start_gc(struct rt6_info *rt) |
| 464 | { |
| 465 | if (ip6_fib_timer.expires == 0 && |
| 466 | (rt->rt6i_flags & (RTF_EXPIRES|RTF_CACHE))) |
| 467 | mod_timer(&ip6_fib_timer, jiffies + ip6_rt_gc_interval); |
| 468 | } |
| 469 | |
| 470 | void fib6_force_start_gc(void) |
| 471 | { |
| 472 | if (ip6_fib_timer.expires == 0) |
| 473 | mod_timer(&ip6_fib_timer, jiffies + ip6_rt_gc_interval); |
| 474 | } |
| 475 | |
| 476 | /* |
| 477 | * Add routing information to the routing tree. |
| 478 | * <destination addr>/<source addr> |
| 479 | * with source addr info in sub-trees |
| 480 | */ |
| 481 | |
| 482 | int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nlmsghdr *nlh, void *_rtattr) |
| 483 | { |
| 484 | struct fib6_node *fn; |
| 485 | int err = -ENOMEM; |
| 486 | |
| 487 | fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr), |
| 488 | rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst)); |
| 489 | |
| 490 | if (fn == NULL) |
| 491 | goto out; |
| 492 | |
| 493 | #ifdef CONFIG_IPV6_SUBTREES |
| 494 | if (rt->rt6i_src.plen) { |
| 495 | struct fib6_node *sn; |
| 496 | |
| 497 | if (fn->subtree == NULL) { |
| 498 | struct fib6_node *sfn; |
| 499 | |
| 500 | /* |
| 501 | * Create subtree. |
| 502 | * |
| 503 | * fn[main tree] |
| 504 | * | |
| 505 | * sfn[subtree root] |
| 506 | * \ |
| 507 | * sn[new leaf node] |
| 508 | */ |
| 509 | |
| 510 | /* Create subtree root node */ |
| 511 | sfn = node_alloc(); |
| 512 | if (sfn == NULL) |
| 513 | goto st_failure; |
| 514 | |
| 515 | sfn->leaf = &ip6_null_entry; |
| 516 | atomic_inc(&ip6_null_entry.rt6i_ref); |
| 517 | sfn->fn_flags = RTN_ROOT; |
| 518 | sfn->fn_sernum = fib6_new_sernum(); |
| 519 | |
| 520 | /* Now add the first leaf node to new subtree */ |
| 521 | |
| 522 | sn = fib6_add_1(sfn, &rt->rt6i_src.addr, |
| 523 | sizeof(struct in6_addr), rt->rt6i_src.plen, |
| 524 | offsetof(struct rt6_info, rt6i_src)); |
| 525 | |
| 526 | if (sn == NULL) { |
| 527 | /* If it is failed, discard just allocated |
| 528 | root, and then (in st_failure) stale node |
| 529 | in main tree. |
| 530 | */ |
| 531 | node_free(sfn); |
| 532 | goto st_failure; |
| 533 | } |
| 534 | |
| 535 | /* Now link new subtree to main tree */ |
| 536 | sfn->parent = fn; |
| 537 | fn->subtree = sfn; |
| 538 | if (fn->leaf == NULL) { |
| 539 | fn->leaf = rt; |
| 540 | atomic_inc(&rt->rt6i_ref); |
| 541 | } |
| 542 | } else { |
| 543 | sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr, |
| 544 | sizeof(struct in6_addr), rt->rt6i_src.plen, |
| 545 | offsetof(struct rt6_info, rt6i_src)); |
| 546 | |
| 547 | if (sn == NULL) |
| 548 | goto st_failure; |
| 549 | } |
| 550 | |
| 551 | fn = sn; |
| 552 | } |
| 553 | #endif |
| 554 | |
| 555 | err = fib6_add_rt2node(fn, rt, nlh); |
| 556 | |
| 557 | if (err == 0) { |
| 558 | fib6_start_gc(rt); |
| 559 | if (!(rt->rt6i_flags&RTF_CACHE)) |
| 560 | fib6_prune_clones(fn, rt); |
| 561 | } |
| 562 | |
| 563 | out: |
| 564 | if (err) |
| 565 | dst_free(&rt->u.dst); |
| 566 | return err; |
| 567 | |
| 568 | #ifdef CONFIG_IPV6_SUBTREES |
| 569 | /* Subtree creation failed, probably main tree node |
| 570 | is orphan. If it is, shoot it. |
| 571 | */ |
| 572 | st_failure: |
| 573 | if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT))) |
| 574 | fib6_repair_tree(fn); |
| 575 | dst_free(&rt->u.dst); |
| 576 | return err; |
| 577 | #endif |
| 578 | } |
| 579 | |
| 580 | /* |
| 581 | * Routing tree lookup |
| 582 | * |
| 583 | */ |
| 584 | |
| 585 | struct lookup_args { |
| 586 | int offset; /* key offset on rt6_info */ |
| 587 | struct in6_addr *addr; /* search key */ |
| 588 | }; |
| 589 | |
| 590 | static struct fib6_node * fib6_lookup_1(struct fib6_node *root, |
| 591 | struct lookup_args *args) |
| 592 | { |
| 593 | struct fib6_node *fn; |
| 594 | int dir; |
| 595 | |
| 596 | /* |
| 597 | * Descend on a tree |
| 598 | */ |
| 599 | |
| 600 | fn = root; |
| 601 | |
| 602 | for (;;) { |
| 603 | struct fib6_node *next; |
| 604 | |
| 605 | dir = addr_bit_set(args->addr, fn->fn_bit); |
| 606 | |
| 607 | next = dir ? fn->right : fn->left; |
| 608 | |
| 609 | if (next) { |
| 610 | fn = next; |
| 611 | continue; |
| 612 | } |
| 613 | |
| 614 | break; |
| 615 | } |
| 616 | |
| 617 | while ((fn->fn_flags & RTN_ROOT) == 0) { |
| 618 | #ifdef CONFIG_IPV6_SUBTREES |
| 619 | if (fn->subtree) { |
| 620 | struct fib6_node *st; |
| 621 | struct lookup_args *narg; |
| 622 | |
| 623 | narg = args + 1; |
| 624 | |
| 625 | if (narg->addr) { |
| 626 | st = fib6_lookup_1(fn->subtree, narg); |
| 627 | |
| 628 | if (st && !(st->fn_flags & RTN_ROOT)) |
| 629 | return st; |
| 630 | } |
| 631 | } |
| 632 | #endif |
| 633 | |
| 634 | if (fn->fn_flags & RTN_RTINFO) { |
| 635 | struct rt6key *key; |
| 636 | |
| 637 | key = (struct rt6key *) ((u8 *) fn->leaf + |
| 638 | args->offset); |
| 639 | |
| 640 | if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) |
| 641 | return fn; |
| 642 | } |
| 643 | |
| 644 | fn = fn->parent; |
| 645 | } |
| 646 | |
| 647 | return NULL; |
| 648 | } |
| 649 | |
| 650 | struct fib6_node * fib6_lookup(struct fib6_node *root, struct in6_addr *daddr, |
| 651 | struct in6_addr *saddr) |
| 652 | { |
| 653 | struct lookup_args args[2]; |
| 654 | struct fib6_node *fn; |
| 655 | |
| 656 | args[0].offset = offsetof(struct rt6_info, rt6i_dst); |
| 657 | args[0].addr = daddr; |
| 658 | |
| 659 | #ifdef CONFIG_IPV6_SUBTREES |
| 660 | args[1].offset = offsetof(struct rt6_info, rt6i_src); |
| 661 | args[1].addr = saddr; |
| 662 | #endif |
| 663 | |
| 664 | fn = fib6_lookup_1(root, args); |
| 665 | |
| 666 | if (fn == NULL || fn->fn_flags & RTN_TL_ROOT) |
| 667 | fn = root; |
| 668 | |
| 669 | return fn; |
| 670 | } |
| 671 | |
| 672 | /* |
| 673 | * Get node with specified destination prefix (and source prefix, |
| 674 | * if subtrees are used) |
| 675 | */ |
| 676 | |
| 677 | |
| 678 | static struct fib6_node * fib6_locate_1(struct fib6_node *root, |
| 679 | struct in6_addr *addr, |
| 680 | int plen, int offset) |
| 681 | { |
| 682 | struct fib6_node *fn; |
| 683 | |
| 684 | for (fn = root; fn ; ) { |
| 685 | struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset); |
| 686 | |
| 687 | /* |
| 688 | * Prefix match |
| 689 | */ |
| 690 | if (plen < fn->fn_bit || |
| 691 | !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) |
| 692 | return NULL; |
| 693 | |
| 694 | if (plen == fn->fn_bit) |
| 695 | return fn; |
| 696 | |
| 697 | /* |
| 698 | * We have more bits to go |
| 699 | */ |
| 700 | if (addr_bit_set(addr, fn->fn_bit)) |
| 701 | fn = fn->right; |
| 702 | else |
| 703 | fn = fn->left; |
| 704 | } |
| 705 | return NULL; |
| 706 | } |
| 707 | |
| 708 | struct fib6_node * fib6_locate(struct fib6_node *root, |
| 709 | struct in6_addr *daddr, int dst_len, |
| 710 | struct in6_addr *saddr, int src_len) |
| 711 | { |
| 712 | struct fib6_node *fn; |
| 713 | |
| 714 | fn = fib6_locate_1(root, daddr, dst_len, |
| 715 | offsetof(struct rt6_info, rt6i_dst)); |
| 716 | |
| 717 | #ifdef CONFIG_IPV6_SUBTREES |
| 718 | if (src_len) { |
| 719 | BUG_TRAP(saddr!=NULL); |
| 720 | if (fn == NULL) |
| 721 | fn = fn->subtree; |
| 722 | if (fn) |
| 723 | fn = fib6_locate_1(fn, saddr, src_len, |
| 724 | offsetof(struct rt6_info, rt6i_src)); |
| 725 | } |
| 726 | #endif |
| 727 | |
| 728 | if (fn && fn->fn_flags&RTN_RTINFO) |
| 729 | return fn; |
| 730 | |
| 731 | return NULL; |
| 732 | } |
| 733 | |
| 734 | |
| 735 | /* |
| 736 | * Deletion |
| 737 | * |
| 738 | */ |
| 739 | |
| 740 | static struct rt6_info * fib6_find_prefix(struct fib6_node *fn) |
| 741 | { |
| 742 | if (fn->fn_flags&RTN_ROOT) |
| 743 | return &ip6_null_entry; |
| 744 | |
| 745 | while(fn) { |
| 746 | if(fn->left) |
| 747 | return fn->left->leaf; |
| 748 | |
| 749 | if(fn->right) |
| 750 | return fn->right->leaf; |
| 751 | |
| 752 | fn = SUBTREE(fn); |
| 753 | } |
| 754 | return NULL; |
| 755 | } |
| 756 | |
| 757 | /* |
| 758 | * Called to trim the tree of intermediate nodes when possible. "fn" |
| 759 | * is the node we want to try and remove. |
| 760 | */ |
| 761 | |
| 762 | static struct fib6_node * fib6_repair_tree(struct fib6_node *fn) |
| 763 | { |
| 764 | int children; |
| 765 | int nstate; |
| 766 | struct fib6_node *child, *pn; |
| 767 | struct fib6_walker_t *w; |
| 768 | int iter = 0; |
| 769 | |
| 770 | for (;;) { |
| 771 | RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter); |
| 772 | iter++; |
| 773 | |
| 774 | BUG_TRAP(!(fn->fn_flags&RTN_RTINFO)); |
| 775 | BUG_TRAP(!(fn->fn_flags&RTN_TL_ROOT)); |
| 776 | BUG_TRAP(fn->leaf==NULL); |
| 777 | |
| 778 | children = 0; |
| 779 | child = NULL; |
| 780 | if (fn->right) child = fn->right, children |= 1; |
| 781 | if (fn->left) child = fn->left, children |= 2; |
| 782 | |
| 783 | if (children == 3 || SUBTREE(fn) |
| 784 | #ifdef CONFIG_IPV6_SUBTREES |
| 785 | /* Subtree root (i.e. fn) may have one child */ |
| 786 | || (children && fn->fn_flags&RTN_ROOT) |
| 787 | #endif |
| 788 | ) { |
| 789 | fn->leaf = fib6_find_prefix(fn); |
| 790 | #if RT6_DEBUG >= 2 |
| 791 | if (fn->leaf==NULL) { |
| 792 | BUG_TRAP(fn->leaf); |
| 793 | fn->leaf = &ip6_null_entry; |
| 794 | } |
| 795 | #endif |
| 796 | atomic_inc(&fn->leaf->rt6i_ref); |
| 797 | return fn->parent; |
| 798 | } |
| 799 | |
| 800 | pn = fn->parent; |
| 801 | #ifdef CONFIG_IPV6_SUBTREES |
| 802 | if (SUBTREE(pn) == fn) { |
| 803 | BUG_TRAP(fn->fn_flags&RTN_ROOT); |
| 804 | SUBTREE(pn) = NULL; |
| 805 | nstate = FWS_L; |
| 806 | } else { |
| 807 | BUG_TRAP(!(fn->fn_flags&RTN_ROOT)); |
| 808 | #endif |
| 809 | if (pn->right == fn) pn->right = child; |
| 810 | else if (pn->left == fn) pn->left = child; |
| 811 | #if RT6_DEBUG >= 2 |
| 812 | else BUG_TRAP(0); |
| 813 | #endif |
| 814 | if (child) |
| 815 | child->parent = pn; |
| 816 | nstate = FWS_R; |
| 817 | #ifdef CONFIG_IPV6_SUBTREES |
| 818 | } |
| 819 | #endif |
| 820 | |
| 821 | read_lock(&fib6_walker_lock); |
| 822 | FOR_WALKERS(w) { |
| 823 | if (child == NULL) { |
| 824 | if (w->root == fn) { |
| 825 | w->root = w->node = NULL; |
| 826 | RT6_TRACE("W %p adjusted by delroot 1\n", w); |
| 827 | } else if (w->node == fn) { |
| 828 | RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate); |
| 829 | w->node = pn; |
| 830 | w->state = nstate; |
| 831 | } |
| 832 | } else { |
| 833 | if (w->root == fn) { |
| 834 | w->root = child; |
| 835 | RT6_TRACE("W %p adjusted by delroot 2\n", w); |
| 836 | } |
| 837 | if (w->node == fn) { |
| 838 | w->node = child; |
| 839 | if (children&2) { |
| 840 | RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); |
| 841 | w->state = w->state>=FWS_R ? FWS_U : FWS_INIT; |
| 842 | } else { |
| 843 | RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state); |
| 844 | w->state = w->state>=FWS_C ? FWS_U : FWS_INIT; |
| 845 | } |
| 846 | } |
| 847 | } |
| 848 | } |
| 849 | read_unlock(&fib6_walker_lock); |
| 850 | |
| 851 | node_free(fn); |
| 852 | if (pn->fn_flags&RTN_RTINFO || SUBTREE(pn)) |
| 853 | return pn; |
| 854 | |
| 855 | rt6_release(pn->leaf); |
| 856 | pn->leaf = NULL; |
| 857 | fn = pn; |
| 858 | } |
| 859 | } |
| 860 | |
| 861 | static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp, |
| 862 | struct nlmsghdr *nlh, void *_rtattr) |
| 863 | { |
| 864 | struct fib6_walker_t *w; |
| 865 | struct rt6_info *rt = *rtp; |
| 866 | |
| 867 | RT6_TRACE("fib6_del_route\n"); |
| 868 | |
| 869 | /* Unlink it */ |
| 870 | *rtp = rt->u.next; |
| 871 | rt->rt6i_node = NULL; |
| 872 | rt6_stats.fib_rt_entries--; |
| 873 | rt6_stats.fib_discarded_routes++; |
| 874 | |
| 875 | /* Adjust walkers */ |
| 876 | read_lock(&fib6_walker_lock); |
| 877 | FOR_WALKERS(w) { |
| 878 | if (w->state == FWS_C && w->leaf == rt) { |
| 879 | RT6_TRACE("walker %p adjusted by delroute\n", w); |
| 880 | w->leaf = rt->u.next; |
| 881 | if (w->leaf == NULL) |
| 882 | w->state = FWS_U; |
| 883 | } |
| 884 | } |
| 885 | read_unlock(&fib6_walker_lock); |
| 886 | |
| 887 | rt->u.next = NULL; |
| 888 | |
| 889 | if (fn->leaf == NULL && fn->fn_flags&RTN_TL_ROOT) |
| 890 | fn->leaf = &ip6_null_entry; |
| 891 | |
| 892 | /* If it was last route, expunge its radix tree node */ |
| 893 | if (fn->leaf == NULL) { |
| 894 | fn->fn_flags &= ~RTN_RTINFO; |
| 895 | rt6_stats.fib_route_nodes--; |
| 896 | fn = fib6_repair_tree(fn); |
| 897 | } |
| 898 | |
| 899 | if (atomic_read(&rt->rt6i_ref) != 1) { |
| 900 | /* This route is used as dummy address holder in some split |
| 901 | * nodes. It is not leaked, but it still holds other resources, |
| 902 | * which must be released in time. So, scan ascendant nodes |
| 903 | * and replace dummy references to this route with references |
| 904 | * to still alive ones. |
| 905 | */ |
| 906 | while (fn) { |
| 907 | if (!(fn->fn_flags&RTN_RTINFO) && fn->leaf == rt) { |
| 908 | fn->leaf = fib6_find_prefix(fn); |
| 909 | atomic_inc(&fn->leaf->rt6i_ref); |
| 910 | rt6_release(rt); |
| 911 | } |
| 912 | fn = fn->parent; |
| 913 | } |
| 914 | /* No more references are possible at this point. */ |
| 915 | if (atomic_read(&rt->rt6i_ref) != 1) BUG(); |
| 916 | } |
| 917 | |
| 918 | inet6_rt_notify(RTM_DELROUTE, rt, nlh); |
| 919 | rt6_release(rt); |
| 920 | } |
| 921 | |
| 922 | int fib6_del(struct rt6_info *rt, struct nlmsghdr *nlh, void *_rtattr) |
| 923 | { |
| 924 | struct fib6_node *fn = rt->rt6i_node; |
| 925 | struct rt6_info **rtp; |
| 926 | |
| 927 | #if RT6_DEBUG >= 2 |
| 928 | if (rt->u.dst.obsolete>0) { |
| 929 | BUG_TRAP(fn==NULL); |
| 930 | return -ENOENT; |
| 931 | } |
| 932 | #endif |
| 933 | if (fn == NULL || rt == &ip6_null_entry) |
| 934 | return -ENOENT; |
| 935 | |
| 936 | BUG_TRAP(fn->fn_flags&RTN_RTINFO); |
| 937 | |
| 938 | if (!(rt->rt6i_flags&RTF_CACHE)) |
| 939 | fib6_prune_clones(fn, rt); |
| 940 | |
| 941 | /* |
| 942 | * Walk the leaf entries looking for ourself |
| 943 | */ |
| 944 | |
| 945 | for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->u.next) { |
| 946 | if (*rtp == rt) { |
| 947 | fib6_del_route(fn, rtp, nlh, _rtattr); |
| 948 | return 0; |
| 949 | } |
| 950 | } |
| 951 | return -ENOENT; |
| 952 | } |
| 953 | |
| 954 | /* |
| 955 | * Tree traversal function. |
| 956 | * |
| 957 | * Certainly, it is not interrupt safe. |
| 958 | * However, it is internally reenterable wrt itself and fib6_add/fib6_del. |
| 959 | * It means, that we can modify tree during walking |
| 960 | * and use this function for garbage collection, clone pruning, |
| 961 | * cleaning tree when a device goes down etc. etc. |
| 962 | * |
| 963 | * It guarantees that every node will be traversed, |
| 964 | * and that it will be traversed only once. |
| 965 | * |
| 966 | * Callback function w->func may return: |
| 967 | * 0 -> continue walking. |
| 968 | * positive value -> walking is suspended (used by tree dumps, |
| 969 | * and probably by gc, if it will be split to several slices) |
| 970 | * negative value -> terminate walking. |
| 971 | * |
| 972 | * The function itself returns: |
| 973 | * 0 -> walk is complete. |
| 974 | * >0 -> walk is incomplete (i.e. suspended) |
| 975 | * <0 -> walk is terminated by an error. |
| 976 | */ |
| 977 | |
| 978 | int fib6_walk_continue(struct fib6_walker_t *w) |
| 979 | { |
| 980 | struct fib6_node *fn, *pn; |
| 981 | |
| 982 | for (;;) { |
| 983 | fn = w->node; |
| 984 | if (fn == NULL) |
| 985 | return 0; |
| 986 | |
| 987 | if (w->prune && fn != w->root && |
| 988 | fn->fn_flags&RTN_RTINFO && w->state < FWS_C) { |
| 989 | w->state = FWS_C; |
| 990 | w->leaf = fn->leaf; |
| 991 | } |
| 992 | switch (w->state) { |
| 993 | #ifdef CONFIG_IPV6_SUBTREES |
| 994 | case FWS_S: |
| 995 | if (SUBTREE(fn)) { |
| 996 | w->node = SUBTREE(fn); |
| 997 | continue; |
| 998 | } |
| 999 | w->state = FWS_L; |
| 1000 | #endif |
| 1001 | case FWS_L: |
| 1002 | if (fn->left) { |
| 1003 | w->node = fn->left; |
| 1004 | w->state = FWS_INIT; |
| 1005 | continue; |
| 1006 | } |
| 1007 | w->state = FWS_R; |
| 1008 | case FWS_R: |
| 1009 | if (fn->right) { |
| 1010 | w->node = fn->right; |
| 1011 | w->state = FWS_INIT; |
| 1012 | continue; |
| 1013 | } |
| 1014 | w->state = FWS_C; |
| 1015 | w->leaf = fn->leaf; |
| 1016 | case FWS_C: |
| 1017 | if (w->leaf && fn->fn_flags&RTN_RTINFO) { |
| 1018 | int err = w->func(w); |
| 1019 | if (err) |
| 1020 | return err; |
| 1021 | continue; |
| 1022 | } |
| 1023 | w->state = FWS_U; |
| 1024 | case FWS_U: |
| 1025 | if (fn == w->root) |
| 1026 | return 0; |
| 1027 | pn = fn->parent; |
| 1028 | w->node = pn; |
| 1029 | #ifdef CONFIG_IPV6_SUBTREES |
| 1030 | if (SUBTREE(pn) == fn) { |
| 1031 | BUG_TRAP(fn->fn_flags&RTN_ROOT); |
| 1032 | w->state = FWS_L; |
| 1033 | continue; |
| 1034 | } |
| 1035 | #endif |
| 1036 | if (pn->left == fn) { |
| 1037 | w->state = FWS_R; |
| 1038 | continue; |
| 1039 | } |
| 1040 | if (pn->right == fn) { |
| 1041 | w->state = FWS_C; |
| 1042 | w->leaf = w->node->leaf; |
| 1043 | continue; |
| 1044 | } |
| 1045 | #if RT6_DEBUG >= 2 |
| 1046 | BUG_TRAP(0); |
| 1047 | #endif |
| 1048 | } |
| 1049 | } |
| 1050 | } |
| 1051 | |
| 1052 | int fib6_walk(struct fib6_walker_t *w) |
| 1053 | { |
| 1054 | int res; |
| 1055 | |
| 1056 | w->state = FWS_INIT; |
| 1057 | w->node = w->root; |
| 1058 | |
| 1059 | fib6_walker_link(w); |
| 1060 | res = fib6_walk_continue(w); |
| 1061 | if (res <= 0) |
| 1062 | fib6_walker_unlink(w); |
| 1063 | return res; |
| 1064 | } |
| 1065 | |
| 1066 | static int fib6_clean_node(struct fib6_walker_t *w) |
| 1067 | { |
| 1068 | int res; |
| 1069 | struct rt6_info *rt; |
| 1070 | struct fib6_cleaner_t *c = (struct fib6_cleaner_t*)w; |
| 1071 | |
| 1072 | for (rt = w->leaf; rt; rt = rt->u.next) { |
| 1073 | res = c->func(rt, c->arg); |
| 1074 | if (res < 0) { |
| 1075 | w->leaf = rt; |
| 1076 | res = fib6_del(rt, NULL, NULL); |
| 1077 | if (res) { |
| 1078 | #if RT6_DEBUG >= 2 |
| 1079 | printk(KERN_DEBUG "fib6_clean_node: del failed: rt=%p@%p err=%d\n", rt, rt->rt6i_node, res); |
| 1080 | #endif |
| 1081 | continue; |
| 1082 | } |
| 1083 | return 0; |
| 1084 | } |
| 1085 | BUG_TRAP(res==0); |
| 1086 | } |
| 1087 | w->leaf = rt; |
| 1088 | return 0; |
| 1089 | } |
| 1090 | |
| 1091 | /* |
| 1092 | * Convenient frontend to tree walker. |
| 1093 | * |
| 1094 | * func is called on each route. |
| 1095 | * It may return -1 -> delete this route. |
| 1096 | * 0 -> continue walking |
| 1097 | * |
| 1098 | * prune==1 -> only immediate children of node (certainly, |
| 1099 | * ignoring pure split nodes) will be scanned. |
| 1100 | */ |
| 1101 | |
| 1102 | void fib6_clean_tree(struct fib6_node *root, |
| 1103 | int (*func)(struct rt6_info *, void *arg), |
| 1104 | int prune, void *arg) |
| 1105 | { |
| 1106 | struct fib6_cleaner_t c; |
| 1107 | |
| 1108 | c.w.root = root; |
| 1109 | c.w.func = fib6_clean_node; |
| 1110 | c.w.prune = prune; |
| 1111 | c.func = func; |
| 1112 | c.arg = arg; |
| 1113 | |
| 1114 | fib6_walk(&c.w); |
| 1115 | } |
| 1116 | |
| 1117 | static int fib6_prune_clone(struct rt6_info *rt, void *arg) |
| 1118 | { |
| 1119 | if (rt->rt6i_flags & RTF_CACHE) { |
| 1120 | RT6_TRACE("pruning clone %p\n", rt); |
| 1121 | return -1; |
| 1122 | } |
| 1123 | |
| 1124 | return 0; |
| 1125 | } |
| 1126 | |
| 1127 | static void fib6_prune_clones(struct fib6_node *fn, struct rt6_info *rt) |
| 1128 | { |
| 1129 | fib6_clean_tree(fn, fib6_prune_clone, 1, rt); |
| 1130 | } |
| 1131 | |
| 1132 | /* |
| 1133 | * Garbage collection |
| 1134 | */ |
| 1135 | |
| 1136 | static struct fib6_gc_args |
| 1137 | { |
| 1138 | int timeout; |
| 1139 | int more; |
| 1140 | } gc_args; |
| 1141 | |
| 1142 | static int fib6_age(struct rt6_info *rt, void *arg) |
| 1143 | { |
| 1144 | unsigned long now = jiffies; |
| 1145 | |
| 1146 | /* |
| 1147 | * check addrconf expiration here. |
| 1148 | * Routes are expired even if they are in use. |
| 1149 | * |
| 1150 | * Also age clones. Note, that clones are aged out |
| 1151 | * only if they are not in use now. |
| 1152 | */ |
| 1153 | |
| 1154 | if (rt->rt6i_flags&RTF_EXPIRES && rt->rt6i_expires) { |
| 1155 | if (time_after(now, rt->rt6i_expires)) { |
| 1156 | RT6_TRACE("expiring %p\n", rt); |
| 1157 | rt6_reset_dflt_pointer(rt); |
| 1158 | return -1; |
| 1159 | } |
| 1160 | gc_args.more++; |
| 1161 | } else if (rt->rt6i_flags & RTF_CACHE) { |
| 1162 | if (atomic_read(&rt->u.dst.__refcnt) == 0 && |
| 1163 | time_after_eq(now, rt->u.dst.lastuse + gc_args.timeout)) { |
| 1164 | RT6_TRACE("aging clone %p\n", rt); |
| 1165 | return -1; |
| 1166 | } else if ((rt->rt6i_flags & RTF_GATEWAY) && |
| 1167 | (!(rt->rt6i_nexthop->flags & NTF_ROUTER))) { |
| 1168 | RT6_TRACE("purging route %p via non-router but gateway\n", |
| 1169 | rt); |
| 1170 | return -1; |
| 1171 | } |
| 1172 | gc_args.more++; |
| 1173 | } |
| 1174 | |
| 1175 | return 0; |
| 1176 | } |
| 1177 | |
| 1178 | static DEFINE_SPINLOCK(fib6_gc_lock); |
| 1179 | |
| 1180 | void fib6_run_gc(unsigned long dummy) |
| 1181 | { |
| 1182 | if (dummy != ~0UL) { |
| 1183 | spin_lock_bh(&fib6_gc_lock); |
| 1184 | gc_args.timeout = dummy ? (int)dummy : ip6_rt_gc_interval; |
| 1185 | } else { |
| 1186 | local_bh_disable(); |
| 1187 | if (!spin_trylock(&fib6_gc_lock)) { |
| 1188 | mod_timer(&ip6_fib_timer, jiffies + HZ); |
| 1189 | local_bh_enable(); |
| 1190 | return; |
| 1191 | } |
| 1192 | gc_args.timeout = ip6_rt_gc_interval; |
| 1193 | } |
| 1194 | gc_args.more = 0; |
| 1195 | |
| 1196 | |
| 1197 | write_lock_bh(&rt6_lock); |
| 1198 | ndisc_dst_gc(&gc_args.more); |
| 1199 | fib6_clean_tree(&ip6_routing_table, fib6_age, 0, NULL); |
| 1200 | write_unlock_bh(&rt6_lock); |
| 1201 | |
| 1202 | if (gc_args.more) |
| 1203 | mod_timer(&ip6_fib_timer, jiffies + ip6_rt_gc_interval); |
| 1204 | else { |
| 1205 | del_timer(&ip6_fib_timer); |
| 1206 | ip6_fib_timer.expires = 0; |
| 1207 | } |
| 1208 | spin_unlock_bh(&fib6_gc_lock); |
| 1209 | } |
| 1210 | |
| 1211 | void __init fib6_init(void) |
| 1212 | { |
| 1213 | fib6_node_kmem = kmem_cache_create("fib6_nodes", |
| 1214 | sizeof(struct fib6_node), |
| 1215 | 0, SLAB_HWCACHE_ALIGN, |
| 1216 | NULL, NULL); |
| 1217 | if (!fib6_node_kmem) |
| 1218 | panic("cannot create fib6_nodes cache"); |
| 1219 | } |
| 1220 | |
| 1221 | void fib6_gc_cleanup(void) |
| 1222 | { |
| 1223 | del_timer(&ip6_fib_timer); |
| 1224 | kmem_cache_destroy(fib6_node_kmem); |
| 1225 | } |