Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame^] | 1 | /* Connection state tracking for netfilter. This is separated from, |
| 2 | but required by, the NAT layer; it can also be used by an iptables |
| 3 | extension. */ |
| 4 | |
| 5 | /* (C) 1999-2001 Paul `Rusty' Russell |
| 6 | * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org> |
| 7 | * |
| 8 | * This program is free software; you can redistribute it and/or modify |
| 9 | * it under the terms of the GNU General Public License version 2 as |
| 10 | * published by the Free Software Foundation. |
| 11 | * |
| 12 | * 23 Apr 2001: Harald Welte <laforge@gnumonks.org> |
| 13 | * - new API and handling of conntrack/nat helpers |
| 14 | * - now capable of multiple expectations for one master |
| 15 | * 16 Jul 2002: Harald Welte <laforge@gnumonks.org> |
| 16 | * - add usage/reference counts to ip_conntrack_expect |
| 17 | * - export ip_conntrack[_expect]_{find_get,put} functions |
| 18 | * */ |
| 19 | |
| 20 | #include <linux/config.h> |
| 21 | #include <linux/types.h> |
| 22 | #include <linux/icmp.h> |
| 23 | #include <linux/ip.h> |
| 24 | #include <linux/netfilter.h> |
| 25 | #include <linux/netfilter_ipv4.h> |
| 26 | #include <linux/module.h> |
| 27 | #include <linux/skbuff.h> |
| 28 | #include <linux/proc_fs.h> |
| 29 | #include <linux/vmalloc.h> |
| 30 | #include <net/checksum.h> |
| 31 | #include <net/ip.h> |
| 32 | #include <linux/stddef.h> |
| 33 | #include <linux/sysctl.h> |
| 34 | #include <linux/slab.h> |
| 35 | #include <linux/random.h> |
| 36 | #include <linux/jhash.h> |
| 37 | #include <linux/err.h> |
| 38 | #include <linux/percpu.h> |
| 39 | #include <linux/moduleparam.h> |
| 40 | |
| 41 | /* This rwlock protects the main hash table, protocol/helper/expected |
| 42 | registrations, conntrack timers*/ |
| 43 | #define ASSERT_READ_LOCK(x) MUST_BE_READ_LOCKED(&ip_conntrack_lock) |
| 44 | #define ASSERT_WRITE_LOCK(x) MUST_BE_WRITE_LOCKED(&ip_conntrack_lock) |
| 45 | |
| 46 | #include <linux/netfilter_ipv4/ip_conntrack.h> |
| 47 | #include <linux/netfilter_ipv4/ip_conntrack_protocol.h> |
| 48 | #include <linux/netfilter_ipv4/ip_conntrack_helper.h> |
| 49 | #include <linux/netfilter_ipv4/ip_conntrack_core.h> |
| 50 | #include <linux/netfilter_ipv4/listhelp.h> |
| 51 | |
| 52 | #define IP_CONNTRACK_VERSION "2.1" |
| 53 | |
| 54 | #if 0 |
| 55 | #define DEBUGP printk |
| 56 | #else |
| 57 | #define DEBUGP(format, args...) |
| 58 | #endif |
| 59 | |
| 60 | DECLARE_RWLOCK(ip_conntrack_lock); |
| 61 | |
| 62 | /* ip_conntrack_standalone needs this */ |
| 63 | atomic_t ip_conntrack_count = ATOMIC_INIT(0); |
| 64 | |
| 65 | void (*ip_conntrack_destroyed)(struct ip_conntrack *conntrack) = NULL; |
| 66 | LIST_HEAD(ip_conntrack_expect_list); |
| 67 | struct ip_conntrack_protocol *ip_ct_protos[MAX_IP_CT_PROTO]; |
| 68 | static LIST_HEAD(helpers); |
| 69 | unsigned int ip_conntrack_htable_size = 0; |
| 70 | int ip_conntrack_max; |
| 71 | struct list_head *ip_conntrack_hash; |
| 72 | static kmem_cache_t *ip_conntrack_cachep; |
| 73 | static kmem_cache_t *ip_conntrack_expect_cachep; |
| 74 | struct ip_conntrack ip_conntrack_untracked; |
| 75 | unsigned int ip_ct_log_invalid; |
| 76 | static LIST_HEAD(unconfirmed); |
| 77 | static int ip_conntrack_vmalloc; |
| 78 | |
| 79 | DEFINE_PER_CPU(struct ip_conntrack_stat, ip_conntrack_stat); |
| 80 | |
| 81 | void |
| 82 | ip_conntrack_put(struct ip_conntrack *ct) |
| 83 | { |
| 84 | IP_NF_ASSERT(ct); |
| 85 | nf_conntrack_put(&ct->ct_general); |
| 86 | } |
| 87 | |
| 88 | static int ip_conntrack_hash_rnd_initted; |
| 89 | static unsigned int ip_conntrack_hash_rnd; |
| 90 | |
| 91 | static u_int32_t |
| 92 | hash_conntrack(const struct ip_conntrack_tuple *tuple) |
| 93 | { |
| 94 | #if 0 |
| 95 | dump_tuple(tuple); |
| 96 | #endif |
| 97 | return (jhash_3words(tuple->src.ip, |
| 98 | (tuple->dst.ip ^ tuple->dst.protonum), |
| 99 | (tuple->src.u.all | (tuple->dst.u.all << 16)), |
| 100 | ip_conntrack_hash_rnd) % ip_conntrack_htable_size); |
| 101 | } |
| 102 | |
| 103 | int |
| 104 | ip_ct_get_tuple(const struct iphdr *iph, |
| 105 | const struct sk_buff *skb, |
| 106 | unsigned int dataoff, |
| 107 | struct ip_conntrack_tuple *tuple, |
| 108 | const struct ip_conntrack_protocol *protocol) |
| 109 | { |
| 110 | /* Never happen */ |
| 111 | if (iph->frag_off & htons(IP_OFFSET)) { |
| 112 | printk("ip_conntrack_core: Frag of proto %u.\n", |
| 113 | iph->protocol); |
| 114 | return 0; |
| 115 | } |
| 116 | |
| 117 | tuple->src.ip = iph->saddr; |
| 118 | tuple->dst.ip = iph->daddr; |
| 119 | tuple->dst.protonum = iph->protocol; |
| 120 | tuple->dst.dir = IP_CT_DIR_ORIGINAL; |
| 121 | |
| 122 | return protocol->pkt_to_tuple(skb, dataoff, tuple); |
| 123 | } |
| 124 | |
| 125 | int |
| 126 | ip_ct_invert_tuple(struct ip_conntrack_tuple *inverse, |
| 127 | const struct ip_conntrack_tuple *orig, |
| 128 | const struct ip_conntrack_protocol *protocol) |
| 129 | { |
| 130 | inverse->src.ip = orig->dst.ip; |
| 131 | inverse->dst.ip = orig->src.ip; |
| 132 | inverse->dst.protonum = orig->dst.protonum; |
| 133 | inverse->dst.dir = !orig->dst.dir; |
| 134 | |
| 135 | return protocol->invert_tuple(inverse, orig); |
| 136 | } |
| 137 | |
| 138 | |
| 139 | /* ip_conntrack_expect helper functions */ |
| 140 | static void destroy_expect(struct ip_conntrack_expect *exp) |
| 141 | { |
| 142 | ip_conntrack_put(exp->master); |
| 143 | IP_NF_ASSERT(!timer_pending(&exp->timeout)); |
| 144 | kmem_cache_free(ip_conntrack_expect_cachep, exp); |
| 145 | CONNTRACK_STAT_INC(expect_delete); |
| 146 | } |
| 147 | |
| 148 | static void unlink_expect(struct ip_conntrack_expect *exp) |
| 149 | { |
| 150 | MUST_BE_WRITE_LOCKED(&ip_conntrack_lock); |
| 151 | list_del(&exp->list); |
| 152 | /* Logically in destroy_expect, but we hold the lock here. */ |
| 153 | exp->master->expecting--; |
| 154 | } |
| 155 | |
| 156 | static void expectation_timed_out(unsigned long ul_expect) |
| 157 | { |
| 158 | struct ip_conntrack_expect *exp = (void *)ul_expect; |
| 159 | |
| 160 | WRITE_LOCK(&ip_conntrack_lock); |
| 161 | unlink_expect(exp); |
| 162 | WRITE_UNLOCK(&ip_conntrack_lock); |
| 163 | destroy_expect(exp); |
| 164 | } |
| 165 | |
| 166 | /* If an expectation for this connection is found, it gets delete from |
| 167 | * global list then returned. */ |
| 168 | static struct ip_conntrack_expect * |
| 169 | find_expectation(const struct ip_conntrack_tuple *tuple) |
| 170 | { |
| 171 | struct ip_conntrack_expect *i; |
| 172 | |
| 173 | list_for_each_entry(i, &ip_conntrack_expect_list, list) { |
| 174 | /* If master is not in hash table yet (ie. packet hasn't left |
| 175 | this machine yet), how can other end know about expected? |
| 176 | Hence these are not the droids you are looking for (if |
| 177 | master ct never got confirmed, we'd hold a reference to it |
| 178 | and weird things would happen to future packets). */ |
| 179 | if (ip_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask) |
| 180 | && is_confirmed(i->master) |
| 181 | && del_timer(&i->timeout)) { |
| 182 | unlink_expect(i); |
| 183 | return i; |
| 184 | } |
| 185 | } |
| 186 | return NULL; |
| 187 | } |
| 188 | |
| 189 | /* delete all expectations for this conntrack */ |
| 190 | static void remove_expectations(struct ip_conntrack *ct) |
| 191 | { |
| 192 | struct ip_conntrack_expect *i, *tmp; |
| 193 | |
| 194 | /* Optimization: most connection never expect any others. */ |
| 195 | if (ct->expecting == 0) |
| 196 | return; |
| 197 | |
| 198 | list_for_each_entry_safe(i, tmp, &ip_conntrack_expect_list, list) { |
| 199 | if (i->master == ct && del_timer(&i->timeout)) { |
| 200 | unlink_expect(i); |
| 201 | destroy_expect(i); |
| 202 | } |
| 203 | } |
| 204 | } |
| 205 | |
| 206 | static void |
| 207 | clean_from_lists(struct ip_conntrack *ct) |
| 208 | { |
| 209 | unsigned int ho, hr; |
| 210 | |
| 211 | DEBUGP("clean_from_lists(%p)\n", ct); |
| 212 | MUST_BE_WRITE_LOCKED(&ip_conntrack_lock); |
| 213 | |
| 214 | ho = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); |
| 215 | hr = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); |
| 216 | LIST_DELETE(&ip_conntrack_hash[ho], &ct->tuplehash[IP_CT_DIR_ORIGINAL]); |
| 217 | LIST_DELETE(&ip_conntrack_hash[hr], &ct->tuplehash[IP_CT_DIR_REPLY]); |
| 218 | |
| 219 | /* Destroy all pending expectations */ |
| 220 | remove_expectations(ct); |
| 221 | } |
| 222 | |
| 223 | static void |
| 224 | destroy_conntrack(struct nf_conntrack *nfct) |
| 225 | { |
| 226 | struct ip_conntrack *ct = (struct ip_conntrack *)nfct; |
| 227 | struct ip_conntrack_protocol *proto; |
| 228 | |
| 229 | DEBUGP("destroy_conntrack(%p)\n", ct); |
| 230 | IP_NF_ASSERT(atomic_read(&nfct->use) == 0); |
| 231 | IP_NF_ASSERT(!timer_pending(&ct->timeout)); |
| 232 | |
| 233 | /* To make sure we don't get any weird locking issues here: |
| 234 | * destroy_conntrack() MUST NOT be called with a write lock |
| 235 | * to ip_conntrack_lock!!! -HW */ |
| 236 | proto = ip_ct_find_proto(ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.protonum); |
| 237 | if (proto && proto->destroy) |
| 238 | proto->destroy(ct); |
| 239 | |
| 240 | if (ip_conntrack_destroyed) |
| 241 | ip_conntrack_destroyed(ct); |
| 242 | |
| 243 | WRITE_LOCK(&ip_conntrack_lock); |
| 244 | /* Expectations will have been removed in clean_from_lists, |
| 245 | * except TFTP can create an expectation on the first packet, |
| 246 | * before connection is in the list, so we need to clean here, |
| 247 | * too. */ |
| 248 | remove_expectations(ct); |
| 249 | |
| 250 | /* We overload first tuple to link into unconfirmed list. */ |
| 251 | if (!is_confirmed(ct)) { |
| 252 | BUG_ON(list_empty(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list)); |
| 253 | list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list); |
| 254 | } |
| 255 | |
| 256 | CONNTRACK_STAT_INC(delete); |
| 257 | WRITE_UNLOCK(&ip_conntrack_lock); |
| 258 | |
| 259 | if (ct->master) |
| 260 | ip_conntrack_put(ct->master); |
| 261 | |
| 262 | DEBUGP("destroy_conntrack: returning ct=%p to slab\n", ct); |
| 263 | kmem_cache_free(ip_conntrack_cachep, ct); |
| 264 | atomic_dec(&ip_conntrack_count); |
| 265 | } |
| 266 | |
| 267 | static void death_by_timeout(unsigned long ul_conntrack) |
| 268 | { |
| 269 | struct ip_conntrack *ct = (void *)ul_conntrack; |
| 270 | |
| 271 | WRITE_LOCK(&ip_conntrack_lock); |
| 272 | /* Inside lock so preempt is disabled on module removal path. |
| 273 | * Otherwise we can get spurious warnings. */ |
| 274 | CONNTRACK_STAT_INC(delete_list); |
| 275 | clean_from_lists(ct); |
| 276 | WRITE_UNLOCK(&ip_conntrack_lock); |
| 277 | ip_conntrack_put(ct); |
| 278 | } |
| 279 | |
| 280 | static inline int |
| 281 | conntrack_tuple_cmp(const struct ip_conntrack_tuple_hash *i, |
| 282 | const struct ip_conntrack_tuple *tuple, |
| 283 | const struct ip_conntrack *ignored_conntrack) |
| 284 | { |
| 285 | MUST_BE_READ_LOCKED(&ip_conntrack_lock); |
| 286 | return tuplehash_to_ctrack(i) != ignored_conntrack |
| 287 | && ip_ct_tuple_equal(tuple, &i->tuple); |
| 288 | } |
| 289 | |
| 290 | static struct ip_conntrack_tuple_hash * |
| 291 | __ip_conntrack_find(const struct ip_conntrack_tuple *tuple, |
| 292 | const struct ip_conntrack *ignored_conntrack) |
| 293 | { |
| 294 | struct ip_conntrack_tuple_hash *h; |
| 295 | unsigned int hash = hash_conntrack(tuple); |
| 296 | |
| 297 | MUST_BE_READ_LOCKED(&ip_conntrack_lock); |
| 298 | list_for_each_entry(h, &ip_conntrack_hash[hash], list) { |
| 299 | if (conntrack_tuple_cmp(h, tuple, ignored_conntrack)) { |
| 300 | CONNTRACK_STAT_INC(found); |
| 301 | return h; |
| 302 | } |
| 303 | CONNTRACK_STAT_INC(searched); |
| 304 | } |
| 305 | |
| 306 | return NULL; |
| 307 | } |
| 308 | |
| 309 | /* Find a connection corresponding to a tuple. */ |
| 310 | struct ip_conntrack_tuple_hash * |
| 311 | ip_conntrack_find_get(const struct ip_conntrack_tuple *tuple, |
| 312 | const struct ip_conntrack *ignored_conntrack) |
| 313 | { |
| 314 | struct ip_conntrack_tuple_hash *h; |
| 315 | |
| 316 | READ_LOCK(&ip_conntrack_lock); |
| 317 | h = __ip_conntrack_find(tuple, ignored_conntrack); |
| 318 | if (h) |
| 319 | atomic_inc(&tuplehash_to_ctrack(h)->ct_general.use); |
| 320 | READ_UNLOCK(&ip_conntrack_lock); |
| 321 | |
| 322 | return h; |
| 323 | } |
| 324 | |
| 325 | /* Confirm a connection given skb; places it in hash table */ |
| 326 | int |
| 327 | __ip_conntrack_confirm(struct sk_buff **pskb) |
| 328 | { |
| 329 | unsigned int hash, repl_hash; |
| 330 | struct ip_conntrack *ct; |
| 331 | enum ip_conntrack_info ctinfo; |
| 332 | |
| 333 | ct = ip_conntrack_get(*pskb, &ctinfo); |
| 334 | |
| 335 | /* ipt_REJECT uses ip_conntrack_attach to attach related |
| 336 | ICMP/TCP RST packets in other direction. Actual packet |
| 337 | which created connection will be IP_CT_NEW or for an |
| 338 | expected connection, IP_CT_RELATED. */ |
| 339 | if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) |
| 340 | return NF_ACCEPT; |
| 341 | |
| 342 | hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); |
| 343 | repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); |
| 344 | |
| 345 | /* We're not in hash table, and we refuse to set up related |
| 346 | connections for unconfirmed conns. But packet copies and |
| 347 | REJECT will give spurious warnings here. */ |
| 348 | /* IP_NF_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ |
| 349 | |
| 350 | /* No external references means noone else could have |
| 351 | confirmed us. */ |
| 352 | IP_NF_ASSERT(!is_confirmed(ct)); |
| 353 | DEBUGP("Confirming conntrack %p\n", ct); |
| 354 | |
| 355 | WRITE_LOCK(&ip_conntrack_lock); |
| 356 | |
| 357 | /* See if there's one in the list already, including reverse: |
| 358 | NAT could have grabbed it without realizing, since we're |
| 359 | not in the hash. If there is, we lost race. */ |
| 360 | if (!LIST_FIND(&ip_conntrack_hash[hash], |
| 361 | conntrack_tuple_cmp, |
| 362 | struct ip_conntrack_tuple_hash *, |
| 363 | &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, NULL) |
| 364 | && !LIST_FIND(&ip_conntrack_hash[repl_hash], |
| 365 | conntrack_tuple_cmp, |
| 366 | struct ip_conntrack_tuple_hash *, |
| 367 | &ct->tuplehash[IP_CT_DIR_REPLY].tuple, NULL)) { |
| 368 | /* Remove from unconfirmed list */ |
| 369 | list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list); |
| 370 | |
| 371 | list_prepend(&ip_conntrack_hash[hash], |
| 372 | &ct->tuplehash[IP_CT_DIR_ORIGINAL]); |
| 373 | list_prepend(&ip_conntrack_hash[repl_hash], |
| 374 | &ct->tuplehash[IP_CT_DIR_REPLY]); |
| 375 | /* Timer relative to confirmation time, not original |
| 376 | setting time, otherwise we'd get timer wrap in |
| 377 | weird delay cases. */ |
| 378 | ct->timeout.expires += jiffies; |
| 379 | add_timer(&ct->timeout); |
| 380 | atomic_inc(&ct->ct_general.use); |
| 381 | set_bit(IPS_CONFIRMED_BIT, &ct->status); |
| 382 | CONNTRACK_STAT_INC(insert); |
| 383 | WRITE_UNLOCK(&ip_conntrack_lock); |
| 384 | return NF_ACCEPT; |
| 385 | } |
| 386 | |
| 387 | CONNTRACK_STAT_INC(insert_failed); |
| 388 | WRITE_UNLOCK(&ip_conntrack_lock); |
| 389 | |
| 390 | return NF_DROP; |
| 391 | } |
| 392 | |
| 393 | /* Returns true if a connection correspondings to the tuple (required |
| 394 | for NAT). */ |
| 395 | int |
| 396 | ip_conntrack_tuple_taken(const struct ip_conntrack_tuple *tuple, |
| 397 | const struct ip_conntrack *ignored_conntrack) |
| 398 | { |
| 399 | struct ip_conntrack_tuple_hash *h; |
| 400 | |
| 401 | READ_LOCK(&ip_conntrack_lock); |
| 402 | h = __ip_conntrack_find(tuple, ignored_conntrack); |
| 403 | READ_UNLOCK(&ip_conntrack_lock); |
| 404 | |
| 405 | return h != NULL; |
| 406 | } |
| 407 | |
| 408 | /* There's a small race here where we may free a just-assured |
| 409 | connection. Too bad: we're in trouble anyway. */ |
| 410 | static inline int unreplied(const struct ip_conntrack_tuple_hash *i) |
| 411 | { |
| 412 | return !(test_bit(IPS_ASSURED_BIT, &tuplehash_to_ctrack(i)->status)); |
| 413 | } |
| 414 | |
| 415 | static int early_drop(struct list_head *chain) |
| 416 | { |
| 417 | /* Traverse backwards: gives us oldest, which is roughly LRU */ |
| 418 | struct ip_conntrack_tuple_hash *h; |
| 419 | struct ip_conntrack *ct = NULL; |
| 420 | int dropped = 0; |
| 421 | |
| 422 | READ_LOCK(&ip_conntrack_lock); |
| 423 | h = LIST_FIND_B(chain, unreplied, struct ip_conntrack_tuple_hash *); |
| 424 | if (h) { |
| 425 | ct = tuplehash_to_ctrack(h); |
| 426 | atomic_inc(&ct->ct_general.use); |
| 427 | } |
| 428 | READ_UNLOCK(&ip_conntrack_lock); |
| 429 | |
| 430 | if (!ct) |
| 431 | return dropped; |
| 432 | |
| 433 | if (del_timer(&ct->timeout)) { |
| 434 | death_by_timeout((unsigned long)ct); |
| 435 | dropped = 1; |
| 436 | CONNTRACK_STAT_INC(early_drop); |
| 437 | } |
| 438 | ip_conntrack_put(ct); |
| 439 | return dropped; |
| 440 | } |
| 441 | |
| 442 | static inline int helper_cmp(const struct ip_conntrack_helper *i, |
| 443 | const struct ip_conntrack_tuple *rtuple) |
| 444 | { |
| 445 | return ip_ct_tuple_mask_cmp(rtuple, &i->tuple, &i->mask); |
| 446 | } |
| 447 | |
| 448 | static struct ip_conntrack_helper *ip_ct_find_helper(const struct ip_conntrack_tuple *tuple) |
| 449 | { |
| 450 | return LIST_FIND(&helpers, helper_cmp, |
| 451 | struct ip_conntrack_helper *, |
| 452 | tuple); |
| 453 | } |
| 454 | |
| 455 | /* Allocate a new conntrack: we return -ENOMEM if classification |
| 456 | failed due to stress. Otherwise it really is unclassifiable. */ |
| 457 | static struct ip_conntrack_tuple_hash * |
| 458 | init_conntrack(const struct ip_conntrack_tuple *tuple, |
| 459 | struct ip_conntrack_protocol *protocol, |
| 460 | struct sk_buff *skb) |
| 461 | { |
| 462 | struct ip_conntrack *conntrack; |
| 463 | struct ip_conntrack_tuple repl_tuple; |
| 464 | size_t hash; |
| 465 | struct ip_conntrack_expect *exp; |
| 466 | |
| 467 | if (!ip_conntrack_hash_rnd_initted) { |
| 468 | get_random_bytes(&ip_conntrack_hash_rnd, 4); |
| 469 | ip_conntrack_hash_rnd_initted = 1; |
| 470 | } |
| 471 | |
| 472 | hash = hash_conntrack(tuple); |
| 473 | |
| 474 | if (ip_conntrack_max |
| 475 | && atomic_read(&ip_conntrack_count) >= ip_conntrack_max) { |
| 476 | /* Try dropping from this hash chain. */ |
| 477 | if (!early_drop(&ip_conntrack_hash[hash])) { |
| 478 | if (net_ratelimit()) |
| 479 | printk(KERN_WARNING |
| 480 | "ip_conntrack: table full, dropping" |
| 481 | " packet.\n"); |
| 482 | return ERR_PTR(-ENOMEM); |
| 483 | } |
| 484 | } |
| 485 | |
| 486 | if (!ip_ct_invert_tuple(&repl_tuple, tuple, protocol)) { |
| 487 | DEBUGP("Can't invert tuple.\n"); |
| 488 | return NULL; |
| 489 | } |
| 490 | |
| 491 | conntrack = kmem_cache_alloc(ip_conntrack_cachep, GFP_ATOMIC); |
| 492 | if (!conntrack) { |
| 493 | DEBUGP("Can't allocate conntrack.\n"); |
| 494 | return ERR_PTR(-ENOMEM); |
| 495 | } |
| 496 | |
| 497 | memset(conntrack, 0, sizeof(*conntrack)); |
| 498 | atomic_set(&conntrack->ct_general.use, 1); |
| 499 | conntrack->ct_general.destroy = destroy_conntrack; |
| 500 | conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *tuple; |
| 501 | conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = repl_tuple; |
| 502 | if (!protocol->new(conntrack, skb)) { |
| 503 | kmem_cache_free(ip_conntrack_cachep, conntrack); |
| 504 | return NULL; |
| 505 | } |
| 506 | /* Don't set timer yet: wait for confirmation */ |
| 507 | init_timer(&conntrack->timeout); |
| 508 | conntrack->timeout.data = (unsigned long)conntrack; |
| 509 | conntrack->timeout.function = death_by_timeout; |
| 510 | |
| 511 | WRITE_LOCK(&ip_conntrack_lock); |
| 512 | exp = find_expectation(tuple); |
| 513 | |
| 514 | if (exp) { |
| 515 | DEBUGP("conntrack: expectation arrives ct=%p exp=%p\n", |
| 516 | conntrack, exp); |
| 517 | /* Welcome, Mr. Bond. We've been expecting you... */ |
| 518 | __set_bit(IPS_EXPECTED_BIT, &conntrack->status); |
| 519 | conntrack->master = exp->master; |
| 520 | #if CONFIG_IP_NF_CONNTRACK_MARK |
| 521 | conntrack->mark = exp->master->mark; |
| 522 | #endif |
| 523 | nf_conntrack_get(&conntrack->master->ct_general); |
| 524 | CONNTRACK_STAT_INC(expect_new); |
| 525 | } else { |
| 526 | conntrack->helper = ip_ct_find_helper(&repl_tuple); |
| 527 | |
| 528 | CONNTRACK_STAT_INC(new); |
| 529 | } |
| 530 | |
| 531 | /* Overload tuple linked list to put us in unconfirmed list. */ |
| 532 | list_add(&conntrack->tuplehash[IP_CT_DIR_ORIGINAL].list, &unconfirmed); |
| 533 | |
| 534 | atomic_inc(&ip_conntrack_count); |
| 535 | WRITE_UNLOCK(&ip_conntrack_lock); |
| 536 | |
| 537 | if (exp) { |
| 538 | if (exp->expectfn) |
| 539 | exp->expectfn(conntrack, exp); |
| 540 | destroy_expect(exp); |
| 541 | } |
| 542 | |
| 543 | return &conntrack->tuplehash[IP_CT_DIR_ORIGINAL]; |
| 544 | } |
| 545 | |
| 546 | /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ |
| 547 | static inline struct ip_conntrack * |
| 548 | resolve_normal_ct(struct sk_buff *skb, |
| 549 | struct ip_conntrack_protocol *proto, |
| 550 | int *set_reply, |
| 551 | unsigned int hooknum, |
| 552 | enum ip_conntrack_info *ctinfo) |
| 553 | { |
| 554 | struct ip_conntrack_tuple tuple; |
| 555 | struct ip_conntrack_tuple_hash *h; |
| 556 | struct ip_conntrack *ct; |
| 557 | |
| 558 | IP_NF_ASSERT((skb->nh.iph->frag_off & htons(IP_OFFSET)) == 0); |
| 559 | |
| 560 | if (!ip_ct_get_tuple(skb->nh.iph, skb, skb->nh.iph->ihl*4, |
| 561 | &tuple,proto)) |
| 562 | return NULL; |
| 563 | |
| 564 | /* look for tuple match */ |
| 565 | h = ip_conntrack_find_get(&tuple, NULL); |
| 566 | if (!h) { |
| 567 | h = init_conntrack(&tuple, proto, skb); |
| 568 | if (!h) |
| 569 | return NULL; |
| 570 | if (IS_ERR(h)) |
| 571 | return (void *)h; |
| 572 | } |
| 573 | ct = tuplehash_to_ctrack(h); |
| 574 | |
| 575 | /* It exists; we have (non-exclusive) reference. */ |
| 576 | if (DIRECTION(h) == IP_CT_DIR_REPLY) { |
| 577 | *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY; |
| 578 | /* Please set reply bit if this packet OK */ |
| 579 | *set_reply = 1; |
| 580 | } else { |
| 581 | /* Once we've had two way comms, always ESTABLISHED. */ |
| 582 | if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { |
| 583 | DEBUGP("ip_conntrack_in: normal packet for %p\n", |
| 584 | ct); |
| 585 | *ctinfo = IP_CT_ESTABLISHED; |
| 586 | } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { |
| 587 | DEBUGP("ip_conntrack_in: related packet for %p\n", |
| 588 | ct); |
| 589 | *ctinfo = IP_CT_RELATED; |
| 590 | } else { |
| 591 | DEBUGP("ip_conntrack_in: new packet for %p\n", |
| 592 | ct); |
| 593 | *ctinfo = IP_CT_NEW; |
| 594 | } |
| 595 | *set_reply = 0; |
| 596 | } |
| 597 | skb->nfct = &ct->ct_general; |
| 598 | skb->nfctinfo = *ctinfo; |
| 599 | return ct; |
| 600 | } |
| 601 | |
| 602 | /* Netfilter hook itself. */ |
| 603 | unsigned int ip_conntrack_in(unsigned int hooknum, |
| 604 | struct sk_buff **pskb, |
| 605 | const struct net_device *in, |
| 606 | const struct net_device *out, |
| 607 | int (*okfn)(struct sk_buff *)) |
| 608 | { |
| 609 | struct ip_conntrack *ct; |
| 610 | enum ip_conntrack_info ctinfo; |
| 611 | struct ip_conntrack_protocol *proto; |
| 612 | int set_reply; |
| 613 | int ret; |
| 614 | |
| 615 | /* Previously seen (loopback or untracked)? Ignore. */ |
| 616 | if ((*pskb)->nfct) { |
| 617 | CONNTRACK_STAT_INC(ignore); |
| 618 | return NF_ACCEPT; |
| 619 | } |
| 620 | |
| 621 | /* Never happen */ |
| 622 | if ((*pskb)->nh.iph->frag_off & htons(IP_OFFSET)) { |
| 623 | if (net_ratelimit()) { |
| 624 | printk(KERN_ERR "ip_conntrack_in: Frag of proto %u (hook=%u)\n", |
| 625 | (*pskb)->nh.iph->protocol, hooknum); |
| 626 | } |
| 627 | return NF_DROP; |
| 628 | } |
| 629 | |
| 630 | /* FIXME: Do this right please. --RR */ |
| 631 | (*pskb)->nfcache |= NFC_UNKNOWN; |
| 632 | |
| 633 | /* Doesn't cover locally-generated broadcast, so not worth it. */ |
| 634 | #if 0 |
| 635 | /* Ignore broadcast: no `connection'. */ |
| 636 | if ((*pskb)->pkt_type == PACKET_BROADCAST) { |
| 637 | printk("Broadcast packet!\n"); |
| 638 | return NF_ACCEPT; |
| 639 | } else if (((*pskb)->nh.iph->daddr & htonl(0x000000FF)) |
| 640 | == htonl(0x000000FF)) { |
| 641 | printk("Should bcast: %u.%u.%u.%u->%u.%u.%u.%u (sk=%p, ptype=%u)\n", |
| 642 | NIPQUAD((*pskb)->nh.iph->saddr), |
| 643 | NIPQUAD((*pskb)->nh.iph->daddr), |
| 644 | (*pskb)->sk, (*pskb)->pkt_type); |
| 645 | } |
| 646 | #endif |
| 647 | |
| 648 | proto = ip_ct_find_proto((*pskb)->nh.iph->protocol); |
| 649 | |
| 650 | /* It may be an special packet, error, unclean... |
| 651 | * inverse of the return code tells to the netfilter |
| 652 | * core what to do with the packet. */ |
| 653 | if (proto->error != NULL |
| 654 | && (ret = proto->error(*pskb, &ctinfo, hooknum)) <= 0) { |
| 655 | CONNTRACK_STAT_INC(error); |
| 656 | CONNTRACK_STAT_INC(invalid); |
| 657 | return -ret; |
| 658 | } |
| 659 | |
| 660 | if (!(ct = resolve_normal_ct(*pskb, proto,&set_reply,hooknum,&ctinfo))) { |
| 661 | /* Not valid part of a connection */ |
| 662 | CONNTRACK_STAT_INC(invalid); |
| 663 | return NF_ACCEPT; |
| 664 | } |
| 665 | |
| 666 | if (IS_ERR(ct)) { |
| 667 | /* Too stressed to deal. */ |
| 668 | CONNTRACK_STAT_INC(drop); |
| 669 | return NF_DROP; |
| 670 | } |
| 671 | |
| 672 | IP_NF_ASSERT((*pskb)->nfct); |
| 673 | |
| 674 | ret = proto->packet(ct, *pskb, ctinfo); |
| 675 | if (ret < 0) { |
| 676 | /* Invalid: inverse of the return code tells |
| 677 | * the netfilter core what to do*/ |
| 678 | nf_conntrack_put((*pskb)->nfct); |
| 679 | (*pskb)->nfct = NULL; |
| 680 | CONNTRACK_STAT_INC(invalid); |
| 681 | return -ret; |
| 682 | } |
| 683 | |
| 684 | if (set_reply) |
| 685 | set_bit(IPS_SEEN_REPLY_BIT, &ct->status); |
| 686 | |
| 687 | return ret; |
| 688 | } |
| 689 | |
| 690 | int invert_tuplepr(struct ip_conntrack_tuple *inverse, |
| 691 | const struct ip_conntrack_tuple *orig) |
| 692 | { |
| 693 | return ip_ct_invert_tuple(inverse, orig, |
| 694 | ip_ct_find_proto(orig->dst.protonum)); |
| 695 | } |
| 696 | |
| 697 | /* Would two expected things clash? */ |
| 698 | static inline int expect_clash(const struct ip_conntrack_expect *a, |
| 699 | const struct ip_conntrack_expect *b) |
| 700 | { |
| 701 | /* Part covered by intersection of masks must be unequal, |
| 702 | otherwise they clash */ |
| 703 | struct ip_conntrack_tuple intersect_mask |
| 704 | = { { a->mask.src.ip & b->mask.src.ip, |
| 705 | { a->mask.src.u.all & b->mask.src.u.all } }, |
| 706 | { a->mask.dst.ip & b->mask.dst.ip, |
| 707 | { a->mask.dst.u.all & b->mask.dst.u.all }, |
| 708 | a->mask.dst.protonum & b->mask.dst.protonum } }; |
| 709 | |
| 710 | return ip_ct_tuple_mask_cmp(&a->tuple, &b->tuple, &intersect_mask); |
| 711 | } |
| 712 | |
| 713 | static inline int expect_matches(const struct ip_conntrack_expect *a, |
| 714 | const struct ip_conntrack_expect *b) |
| 715 | { |
| 716 | return a->master == b->master |
| 717 | && ip_ct_tuple_equal(&a->tuple, &b->tuple) |
| 718 | && ip_ct_tuple_equal(&a->mask, &b->mask); |
| 719 | } |
| 720 | |
| 721 | /* Generally a bad idea to call this: could have matched already. */ |
| 722 | void ip_conntrack_unexpect_related(struct ip_conntrack_expect *exp) |
| 723 | { |
| 724 | struct ip_conntrack_expect *i; |
| 725 | |
| 726 | WRITE_LOCK(&ip_conntrack_lock); |
| 727 | /* choose the the oldest expectation to evict */ |
| 728 | list_for_each_entry_reverse(i, &ip_conntrack_expect_list, list) { |
| 729 | if (expect_matches(i, exp) && del_timer(&i->timeout)) { |
| 730 | unlink_expect(i); |
| 731 | WRITE_UNLOCK(&ip_conntrack_lock); |
| 732 | destroy_expect(i); |
| 733 | return; |
| 734 | } |
| 735 | } |
| 736 | WRITE_UNLOCK(&ip_conntrack_lock); |
| 737 | } |
| 738 | |
| 739 | struct ip_conntrack_expect *ip_conntrack_expect_alloc(void) |
| 740 | { |
| 741 | struct ip_conntrack_expect *new; |
| 742 | |
| 743 | new = kmem_cache_alloc(ip_conntrack_expect_cachep, GFP_ATOMIC); |
| 744 | if (!new) { |
| 745 | DEBUGP("expect_related: OOM allocating expect\n"); |
| 746 | return NULL; |
| 747 | } |
| 748 | new->master = NULL; |
| 749 | return new; |
| 750 | } |
| 751 | |
| 752 | void ip_conntrack_expect_free(struct ip_conntrack_expect *expect) |
| 753 | { |
| 754 | kmem_cache_free(ip_conntrack_expect_cachep, expect); |
| 755 | } |
| 756 | |
| 757 | static void ip_conntrack_expect_insert(struct ip_conntrack_expect *exp) |
| 758 | { |
| 759 | atomic_inc(&exp->master->ct_general.use); |
| 760 | exp->master->expecting++; |
| 761 | list_add(&exp->list, &ip_conntrack_expect_list); |
| 762 | |
| 763 | if (exp->master->helper->timeout) { |
| 764 | init_timer(&exp->timeout); |
| 765 | exp->timeout.data = (unsigned long)exp; |
| 766 | exp->timeout.function = expectation_timed_out; |
| 767 | exp->timeout.expires |
| 768 | = jiffies + exp->master->helper->timeout * HZ; |
| 769 | add_timer(&exp->timeout); |
| 770 | } else |
| 771 | exp->timeout.function = NULL; |
| 772 | |
| 773 | CONNTRACK_STAT_INC(expect_create); |
| 774 | } |
| 775 | |
| 776 | /* Race with expectations being used means we could have none to find; OK. */ |
| 777 | static void evict_oldest_expect(struct ip_conntrack *master) |
| 778 | { |
| 779 | struct ip_conntrack_expect *i; |
| 780 | |
| 781 | list_for_each_entry_reverse(i, &ip_conntrack_expect_list, list) { |
| 782 | if (i->master == master) { |
| 783 | if (del_timer(&i->timeout)) { |
| 784 | unlink_expect(i); |
| 785 | destroy_expect(i); |
| 786 | } |
| 787 | break; |
| 788 | } |
| 789 | } |
| 790 | } |
| 791 | |
| 792 | static inline int refresh_timer(struct ip_conntrack_expect *i) |
| 793 | { |
| 794 | if (!del_timer(&i->timeout)) |
| 795 | return 0; |
| 796 | |
| 797 | i->timeout.expires = jiffies + i->master->helper->timeout*HZ; |
| 798 | add_timer(&i->timeout); |
| 799 | return 1; |
| 800 | } |
| 801 | |
| 802 | int ip_conntrack_expect_related(struct ip_conntrack_expect *expect) |
| 803 | { |
| 804 | struct ip_conntrack_expect *i; |
| 805 | int ret; |
| 806 | |
| 807 | DEBUGP("ip_conntrack_expect_related %p\n", related_to); |
| 808 | DEBUGP("tuple: "); DUMP_TUPLE(&expect->tuple); |
| 809 | DEBUGP("mask: "); DUMP_TUPLE(&expect->mask); |
| 810 | |
| 811 | WRITE_LOCK(&ip_conntrack_lock); |
| 812 | list_for_each_entry(i, &ip_conntrack_expect_list, list) { |
| 813 | if (expect_matches(i, expect)) { |
| 814 | /* Refresh timer: if it's dying, ignore.. */ |
| 815 | if (refresh_timer(i)) { |
| 816 | ret = 0; |
| 817 | /* We don't need the one they've given us. */ |
| 818 | ip_conntrack_expect_free(expect); |
| 819 | goto out; |
| 820 | } |
| 821 | } else if (expect_clash(i, expect)) { |
| 822 | ret = -EBUSY; |
| 823 | goto out; |
| 824 | } |
| 825 | } |
| 826 | |
| 827 | /* Will be over limit? */ |
| 828 | if (expect->master->helper->max_expected && |
| 829 | expect->master->expecting >= expect->master->helper->max_expected) |
| 830 | evict_oldest_expect(expect->master); |
| 831 | |
| 832 | ip_conntrack_expect_insert(expect); |
| 833 | ret = 0; |
| 834 | out: |
| 835 | WRITE_UNLOCK(&ip_conntrack_lock); |
| 836 | return ret; |
| 837 | } |
| 838 | |
| 839 | /* Alter reply tuple (maybe alter helper). This is for NAT, and is |
| 840 | implicitly racy: see __ip_conntrack_confirm */ |
| 841 | void ip_conntrack_alter_reply(struct ip_conntrack *conntrack, |
| 842 | const struct ip_conntrack_tuple *newreply) |
| 843 | { |
| 844 | WRITE_LOCK(&ip_conntrack_lock); |
| 845 | /* Should be unconfirmed, so not in hash table yet */ |
| 846 | IP_NF_ASSERT(!is_confirmed(conntrack)); |
| 847 | |
| 848 | DEBUGP("Altering reply tuple of %p to ", conntrack); |
| 849 | DUMP_TUPLE(newreply); |
| 850 | |
| 851 | conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; |
| 852 | if (!conntrack->master && conntrack->expecting == 0) |
| 853 | conntrack->helper = ip_ct_find_helper(newreply); |
| 854 | WRITE_UNLOCK(&ip_conntrack_lock); |
| 855 | } |
| 856 | |
| 857 | int ip_conntrack_helper_register(struct ip_conntrack_helper *me) |
| 858 | { |
| 859 | BUG_ON(me->timeout == 0); |
| 860 | WRITE_LOCK(&ip_conntrack_lock); |
| 861 | list_prepend(&helpers, me); |
| 862 | WRITE_UNLOCK(&ip_conntrack_lock); |
| 863 | |
| 864 | return 0; |
| 865 | } |
| 866 | |
| 867 | static inline int unhelp(struct ip_conntrack_tuple_hash *i, |
| 868 | const struct ip_conntrack_helper *me) |
| 869 | { |
| 870 | if (tuplehash_to_ctrack(i)->helper == me) |
| 871 | tuplehash_to_ctrack(i)->helper = NULL; |
| 872 | return 0; |
| 873 | } |
| 874 | |
| 875 | void ip_conntrack_helper_unregister(struct ip_conntrack_helper *me) |
| 876 | { |
| 877 | unsigned int i; |
| 878 | struct ip_conntrack_expect *exp, *tmp; |
| 879 | |
| 880 | /* Need write lock here, to delete helper. */ |
| 881 | WRITE_LOCK(&ip_conntrack_lock); |
| 882 | LIST_DELETE(&helpers, me); |
| 883 | |
| 884 | /* Get rid of expectations */ |
| 885 | list_for_each_entry_safe(exp, tmp, &ip_conntrack_expect_list, list) { |
| 886 | if (exp->master->helper == me && del_timer(&exp->timeout)) { |
| 887 | unlink_expect(exp); |
| 888 | destroy_expect(exp); |
| 889 | } |
| 890 | } |
| 891 | /* Get rid of expecteds, set helpers to NULL. */ |
| 892 | LIST_FIND_W(&unconfirmed, unhelp, struct ip_conntrack_tuple_hash*, me); |
| 893 | for (i = 0; i < ip_conntrack_htable_size; i++) |
| 894 | LIST_FIND_W(&ip_conntrack_hash[i], unhelp, |
| 895 | struct ip_conntrack_tuple_hash *, me); |
| 896 | WRITE_UNLOCK(&ip_conntrack_lock); |
| 897 | |
| 898 | /* Someone could be still looking at the helper in a bh. */ |
| 899 | synchronize_net(); |
| 900 | } |
| 901 | |
| 902 | static inline void ct_add_counters(struct ip_conntrack *ct, |
| 903 | enum ip_conntrack_info ctinfo, |
| 904 | const struct sk_buff *skb) |
| 905 | { |
| 906 | #ifdef CONFIG_IP_NF_CT_ACCT |
| 907 | if (skb) { |
| 908 | ct->counters[CTINFO2DIR(ctinfo)].packets++; |
| 909 | ct->counters[CTINFO2DIR(ctinfo)].bytes += |
| 910 | ntohs(skb->nh.iph->tot_len); |
| 911 | } |
| 912 | #endif |
| 913 | } |
| 914 | |
| 915 | /* Refresh conntrack for this many jiffies and do accounting (if skb != NULL) */ |
| 916 | void ip_ct_refresh_acct(struct ip_conntrack *ct, |
| 917 | enum ip_conntrack_info ctinfo, |
| 918 | const struct sk_buff *skb, |
| 919 | unsigned long extra_jiffies) |
| 920 | { |
| 921 | IP_NF_ASSERT(ct->timeout.data == (unsigned long)ct); |
| 922 | |
| 923 | /* If not in hash table, timer will not be active yet */ |
| 924 | if (!is_confirmed(ct)) { |
| 925 | ct->timeout.expires = extra_jiffies; |
| 926 | ct_add_counters(ct, ctinfo, skb); |
| 927 | } else { |
| 928 | WRITE_LOCK(&ip_conntrack_lock); |
| 929 | /* Need del_timer for race avoidance (may already be dying). */ |
| 930 | if (del_timer(&ct->timeout)) { |
| 931 | ct->timeout.expires = jiffies + extra_jiffies; |
| 932 | add_timer(&ct->timeout); |
| 933 | } |
| 934 | ct_add_counters(ct, ctinfo, skb); |
| 935 | WRITE_UNLOCK(&ip_conntrack_lock); |
| 936 | } |
| 937 | } |
| 938 | |
| 939 | /* Returns new sk_buff, or NULL */ |
| 940 | struct sk_buff * |
| 941 | ip_ct_gather_frags(struct sk_buff *skb, u_int32_t user) |
| 942 | { |
| 943 | struct sock *sk = skb->sk; |
| 944 | #ifdef CONFIG_NETFILTER_DEBUG |
| 945 | unsigned int olddebug = skb->nf_debug; |
| 946 | #endif |
| 947 | |
| 948 | if (sk) { |
| 949 | sock_hold(sk); |
| 950 | skb_orphan(skb); |
| 951 | } |
| 952 | |
| 953 | local_bh_disable(); |
| 954 | skb = ip_defrag(skb, user); |
| 955 | local_bh_enable(); |
| 956 | |
| 957 | if (!skb) { |
| 958 | if (sk) |
| 959 | sock_put(sk); |
| 960 | return skb; |
| 961 | } |
| 962 | |
| 963 | if (sk) { |
| 964 | skb_set_owner_w(skb, sk); |
| 965 | sock_put(sk); |
| 966 | } |
| 967 | |
| 968 | ip_send_check(skb->nh.iph); |
| 969 | skb->nfcache |= NFC_ALTERED; |
| 970 | #ifdef CONFIG_NETFILTER_DEBUG |
| 971 | /* Packet path as if nothing had happened. */ |
| 972 | skb->nf_debug = olddebug; |
| 973 | #endif |
| 974 | return skb; |
| 975 | } |
| 976 | |
| 977 | /* Used by ipt_REJECT. */ |
| 978 | static void ip_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb) |
| 979 | { |
| 980 | struct ip_conntrack *ct; |
| 981 | enum ip_conntrack_info ctinfo; |
| 982 | |
| 983 | /* This ICMP is in reverse direction to the packet which caused it */ |
| 984 | ct = ip_conntrack_get(skb, &ctinfo); |
| 985 | |
| 986 | if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) |
| 987 | ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY; |
| 988 | else |
| 989 | ctinfo = IP_CT_RELATED; |
| 990 | |
| 991 | /* Attach to new skbuff, and increment count */ |
| 992 | nskb->nfct = &ct->ct_general; |
| 993 | nskb->nfctinfo = ctinfo; |
| 994 | nf_conntrack_get(nskb->nfct); |
| 995 | } |
| 996 | |
| 997 | static inline int |
| 998 | do_iter(const struct ip_conntrack_tuple_hash *i, |
| 999 | int (*iter)(struct ip_conntrack *i, void *data), |
| 1000 | void *data) |
| 1001 | { |
| 1002 | return iter(tuplehash_to_ctrack(i), data); |
| 1003 | } |
| 1004 | |
| 1005 | /* Bring out ya dead! */ |
| 1006 | static struct ip_conntrack_tuple_hash * |
| 1007 | get_next_corpse(int (*iter)(struct ip_conntrack *i, void *data), |
| 1008 | void *data, unsigned int *bucket) |
| 1009 | { |
| 1010 | struct ip_conntrack_tuple_hash *h = NULL; |
| 1011 | |
| 1012 | WRITE_LOCK(&ip_conntrack_lock); |
| 1013 | for (; *bucket < ip_conntrack_htable_size; (*bucket)++) { |
| 1014 | h = LIST_FIND_W(&ip_conntrack_hash[*bucket], do_iter, |
| 1015 | struct ip_conntrack_tuple_hash *, iter, data); |
| 1016 | if (h) |
| 1017 | break; |
| 1018 | } |
| 1019 | if (!h) |
| 1020 | h = LIST_FIND_W(&unconfirmed, do_iter, |
| 1021 | struct ip_conntrack_tuple_hash *, iter, data); |
| 1022 | if (h) |
| 1023 | atomic_inc(&tuplehash_to_ctrack(h)->ct_general.use); |
| 1024 | WRITE_UNLOCK(&ip_conntrack_lock); |
| 1025 | |
| 1026 | return h; |
| 1027 | } |
| 1028 | |
| 1029 | void |
| 1030 | ip_ct_iterate_cleanup(int (*iter)(struct ip_conntrack *i, void *), void *data) |
| 1031 | { |
| 1032 | struct ip_conntrack_tuple_hash *h; |
| 1033 | unsigned int bucket = 0; |
| 1034 | |
| 1035 | while ((h = get_next_corpse(iter, data, &bucket)) != NULL) { |
| 1036 | struct ip_conntrack *ct = tuplehash_to_ctrack(h); |
| 1037 | /* Time to push up daises... */ |
| 1038 | if (del_timer(&ct->timeout)) |
| 1039 | death_by_timeout((unsigned long)ct); |
| 1040 | /* ... else the timer will get him soon. */ |
| 1041 | |
| 1042 | ip_conntrack_put(ct); |
| 1043 | } |
| 1044 | } |
| 1045 | |
| 1046 | /* Fast function for those who don't want to parse /proc (and I don't |
| 1047 | blame them). */ |
| 1048 | /* Reversing the socket's dst/src point of view gives us the reply |
| 1049 | mapping. */ |
| 1050 | static int |
| 1051 | getorigdst(struct sock *sk, int optval, void __user *user, int *len) |
| 1052 | { |
| 1053 | struct inet_sock *inet = inet_sk(sk); |
| 1054 | struct ip_conntrack_tuple_hash *h; |
| 1055 | struct ip_conntrack_tuple tuple; |
| 1056 | |
| 1057 | IP_CT_TUPLE_U_BLANK(&tuple); |
| 1058 | tuple.src.ip = inet->rcv_saddr; |
| 1059 | tuple.src.u.tcp.port = inet->sport; |
| 1060 | tuple.dst.ip = inet->daddr; |
| 1061 | tuple.dst.u.tcp.port = inet->dport; |
| 1062 | tuple.dst.protonum = IPPROTO_TCP; |
| 1063 | |
| 1064 | /* We only do TCP at the moment: is there a better way? */ |
| 1065 | if (strcmp(sk->sk_prot->name, "TCP")) { |
| 1066 | DEBUGP("SO_ORIGINAL_DST: Not a TCP socket\n"); |
| 1067 | return -ENOPROTOOPT; |
| 1068 | } |
| 1069 | |
| 1070 | if ((unsigned int) *len < sizeof(struct sockaddr_in)) { |
| 1071 | DEBUGP("SO_ORIGINAL_DST: len %u not %u\n", |
| 1072 | *len, sizeof(struct sockaddr_in)); |
| 1073 | return -EINVAL; |
| 1074 | } |
| 1075 | |
| 1076 | h = ip_conntrack_find_get(&tuple, NULL); |
| 1077 | if (h) { |
| 1078 | struct sockaddr_in sin; |
| 1079 | struct ip_conntrack *ct = tuplehash_to_ctrack(h); |
| 1080 | |
| 1081 | sin.sin_family = AF_INET; |
| 1082 | sin.sin_port = ct->tuplehash[IP_CT_DIR_ORIGINAL] |
| 1083 | .tuple.dst.u.tcp.port; |
| 1084 | sin.sin_addr.s_addr = ct->tuplehash[IP_CT_DIR_ORIGINAL] |
| 1085 | .tuple.dst.ip; |
| 1086 | |
| 1087 | DEBUGP("SO_ORIGINAL_DST: %u.%u.%u.%u %u\n", |
| 1088 | NIPQUAD(sin.sin_addr.s_addr), ntohs(sin.sin_port)); |
| 1089 | ip_conntrack_put(ct); |
| 1090 | if (copy_to_user(user, &sin, sizeof(sin)) != 0) |
| 1091 | return -EFAULT; |
| 1092 | else |
| 1093 | return 0; |
| 1094 | } |
| 1095 | DEBUGP("SO_ORIGINAL_DST: Can't find %u.%u.%u.%u/%u-%u.%u.%u.%u/%u.\n", |
| 1096 | NIPQUAD(tuple.src.ip), ntohs(tuple.src.u.tcp.port), |
| 1097 | NIPQUAD(tuple.dst.ip), ntohs(tuple.dst.u.tcp.port)); |
| 1098 | return -ENOENT; |
| 1099 | } |
| 1100 | |
| 1101 | static struct nf_sockopt_ops so_getorigdst = { |
| 1102 | .pf = PF_INET, |
| 1103 | .get_optmin = SO_ORIGINAL_DST, |
| 1104 | .get_optmax = SO_ORIGINAL_DST+1, |
| 1105 | .get = &getorigdst, |
| 1106 | }; |
| 1107 | |
| 1108 | static int kill_all(struct ip_conntrack *i, void *data) |
| 1109 | { |
| 1110 | return 1; |
| 1111 | } |
| 1112 | |
| 1113 | static void free_conntrack_hash(void) |
| 1114 | { |
| 1115 | if (ip_conntrack_vmalloc) |
| 1116 | vfree(ip_conntrack_hash); |
| 1117 | else |
| 1118 | free_pages((unsigned long)ip_conntrack_hash, |
| 1119 | get_order(sizeof(struct list_head) |
| 1120 | * ip_conntrack_htable_size)); |
| 1121 | } |
| 1122 | |
| 1123 | /* Mishearing the voices in his head, our hero wonders how he's |
| 1124 | supposed to kill the mall. */ |
| 1125 | void ip_conntrack_cleanup(void) |
| 1126 | { |
| 1127 | ip_ct_attach = NULL; |
| 1128 | /* This makes sure all current packets have passed through |
| 1129 | netfilter framework. Roll on, two-stage module |
| 1130 | delete... */ |
| 1131 | synchronize_net(); |
| 1132 | |
| 1133 | i_see_dead_people: |
| 1134 | ip_ct_iterate_cleanup(kill_all, NULL); |
| 1135 | if (atomic_read(&ip_conntrack_count) != 0) { |
| 1136 | schedule(); |
| 1137 | goto i_see_dead_people; |
| 1138 | } |
| 1139 | |
| 1140 | kmem_cache_destroy(ip_conntrack_cachep); |
| 1141 | kmem_cache_destroy(ip_conntrack_expect_cachep); |
| 1142 | free_conntrack_hash(); |
| 1143 | nf_unregister_sockopt(&so_getorigdst); |
| 1144 | } |
| 1145 | |
| 1146 | static int hashsize; |
| 1147 | module_param(hashsize, int, 0400); |
| 1148 | |
| 1149 | int __init ip_conntrack_init(void) |
| 1150 | { |
| 1151 | unsigned int i; |
| 1152 | int ret; |
| 1153 | |
| 1154 | /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB |
| 1155 | * machine has 256 buckets. >= 1GB machines have 8192 buckets. */ |
| 1156 | if (hashsize) { |
| 1157 | ip_conntrack_htable_size = hashsize; |
| 1158 | } else { |
| 1159 | ip_conntrack_htable_size |
| 1160 | = (((num_physpages << PAGE_SHIFT) / 16384) |
| 1161 | / sizeof(struct list_head)); |
| 1162 | if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE)) |
| 1163 | ip_conntrack_htable_size = 8192; |
| 1164 | if (ip_conntrack_htable_size < 16) |
| 1165 | ip_conntrack_htable_size = 16; |
| 1166 | } |
| 1167 | ip_conntrack_max = 8 * ip_conntrack_htable_size; |
| 1168 | |
| 1169 | printk("ip_conntrack version %s (%u buckets, %d max)" |
| 1170 | " - %Zd bytes per conntrack\n", IP_CONNTRACK_VERSION, |
| 1171 | ip_conntrack_htable_size, ip_conntrack_max, |
| 1172 | sizeof(struct ip_conntrack)); |
| 1173 | |
| 1174 | ret = nf_register_sockopt(&so_getorigdst); |
| 1175 | if (ret != 0) { |
| 1176 | printk(KERN_ERR "Unable to register netfilter socket option\n"); |
| 1177 | return ret; |
| 1178 | } |
| 1179 | |
| 1180 | /* AK: the hash table is twice as big than needed because it |
| 1181 | uses list_head. it would be much nicer to caches to use a |
| 1182 | single pointer list head here. */ |
| 1183 | ip_conntrack_vmalloc = 0; |
| 1184 | ip_conntrack_hash |
| 1185 | =(void*)__get_free_pages(GFP_KERNEL, |
| 1186 | get_order(sizeof(struct list_head) |
| 1187 | *ip_conntrack_htable_size)); |
| 1188 | if (!ip_conntrack_hash) { |
| 1189 | ip_conntrack_vmalloc = 1; |
| 1190 | printk(KERN_WARNING "ip_conntrack: falling back to vmalloc.\n"); |
| 1191 | ip_conntrack_hash = vmalloc(sizeof(struct list_head) |
| 1192 | * ip_conntrack_htable_size); |
| 1193 | } |
| 1194 | if (!ip_conntrack_hash) { |
| 1195 | printk(KERN_ERR "Unable to create ip_conntrack_hash\n"); |
| 1196 | goto err_unreg_sockopt; |
| 1197 | } |
| 1198 | |
| 1199 | ip_conntrack_cachep = kmem_cache_create("ip_conntrack", |
| 1200 | sizeof(struct ip_conntrack), 0, |
| 1201 | 0, NULL, NULL); |
| 1202 | if (!ip_conntrack_cachep) { |
| 1203 | printk(KERN_ERR "Unable to create ip_conntrack slab cache\n"); |
| 1204 | goto err_free_hash; |
| 1205 | } |
| 1206 | |
| 1207 | ip_conntrack_expect_cachep = kmem_cache_create("ip_conntrack_expect", |
| 1208 | sizeof(struct ip_conntrack_expect), |
| 1209 | 0, 0, NULL, NULL); |
| 1210 | if (!ip_conntrack_expect_cachep) { |
| 1211 | printk(KERN_ERR "Unable to create ip_expect slab cache\n"); |
| 1212 | goto err_free_conntrack_slab; |
| 1213 | } |
| 1214 | |
| 1215 | /* Don't NEED lock here, but good form anyway. */ |
| 1216 | WRITE_LOCK(&ip_conntrack_lock); |
| 1217 | for (i = 0; i < MAX_IP_CT_PROTO; i++) |
| 1218 | ip_ct_protos[i] = &ip_conntrack_generic_protocol; |
| 1219 | /* Sew in builtin protocols. */ |
| 1220 | ip_ct_protos[IPPROTO_TCP] = &ip_conntrack_protocol_tcp; |
| 1221 | ip_ct_protos[IPPROTO_UDP] = &ip_conntrack_protocol_udp; |
| 1222 | ip_ct_protos[IPPROTO_ICMP] = &ip_conntrack_protocol_icmp; |
| 1223 | WRITE_UNLOCK(&ip_conntrack_lock); |
| 1224 | |
| 1225 | for (i = 0; i < ip_conntrack_htable_size; i++) |
| 1226 | INIT_LIST_HEAD(&ip_conntrack_hash[i]); |
| 1227 | |
| 1228 | /* For use by ipt_REJECT */ |
| 1229 | ip_ct_attach = ip_conntrack_attach; |
| 1230 | |
| 1231 | /* Set up fake conntrack: |
| 1232 | - to never be deleted, not in any hashes */ |
| 1233 | atomic_set(&ip_conntrack_untracked.ct_general.use, 1); |
| 1234 | /* - and look it like as a confirmed connection */ |
| 1235 | set_bit(IPS_CONFIRMED_BIT, &ip_conntrack_untracked.status); |
| 1236 | |
| 1237 | return ret; |
| 1238 | |
| 1239 | err_free_conntrack_slab: |
| 1240 | kmem_cache_destroy(ip_conntrack_cachep); |
| 1241 | err_free_hash: |
| 1242 | free_conntrack_hash(); |
| 1243 | err_unreg_sockopt: |
| 1244 | nf_unregister_sockopt(&so_getorigdst); |
| 1245 | |
| 1246 | return -ENOMEM; |
| 1247 | } |