Martin Josefsson | 77ab9cf | 2006-11-29 02:34:58 +0100 | [diff] [blame] | 1 | /* Expectation handling for nf_conntrack. */ |
| 2 | |
| 3 | /* (C) 1999-2001 Paul `Rusty' Russell |
| 4 | * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org> |
| 5 | * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org> |
| 6 | * |
| 7 | * This program is free software; you can redistribute it and/or modify |
| 8 | * it under the terms of the GNU General Public License version 2 as |
| 9 | * published by the Free Software Foundation. |
| 10 | */ |
| 11 | |
| 12 | #include <linux/types.h> |
| 13 | #include <linux/netfilter.h> |
| 14 | #include <linux/skbuff.h> |
| 15 | #include <linux/proc_fs.h> |
| 16 | #include <linux/seq_file.h> |
| 17 | #include <linux/stddef.h> |
| 18 | #include <linux/slab.h> |
| 19 | #include <linux/err.h> |
| 20 | #include <linux/percpu.h> |
| 21 | #include <linux/kernel.h> |
| 22 | |
| 23 | #include <net/netfilter/nf_conntrack.h> |
| 24 | #include <net/netfilter/nf_conntrack_core.h> |
| 25 | #include <net/netfilter/nf_conntrack_expect.h> |
| 26 | #include <net/netfilter/nf_conntrack_helper.h> |
| 27 | #include <net/netfilter/nf_conntrack_tuple.h> |
| 28 | |
| 29 | LIST_HEAD(nf_conntrack_expect_list); |
| 30 | kmem_cache_t *nf_conntrack_expect_cachep __read_mostly; |
Martin Josefsson | 77ab9cf | 2006-11-29 02:34:58 +0100 | [diff] [blame] | 31 | static unsigned int nf_conntrack_expect_next_id; |
| 32 | |
| 33 | /* nf_conntrack_expect helper functions */ |
| 34 | void nf_ct_unlink_expect(struct nf_conntrack_expect *exp) |
| 35 | { |
| 36 | struct nf_conn_help *master_help = nfct_help(exp->master); |
| 37 | |
| 38 | NF_CT_ASSERT(master_help); |
| 39 | NF_CT_ASSERT(!timer_pending(&exp->timeout)); |
| 40 | |
| 41 | list_del(&exp->list); |
| 42 | NF_CT_STAT_INC(expect_delete); |
| 43 | master_help->expecting--; |
| 44 | nf_conntrack_expect_put(exp); |
| 45 | } |
| 46 | |
| 47 | static void expectation_timed_out(unsigned long ul_expect) |
| 48 | { |
| 49 | struct nf_conntrack_expect *exp = (void *)ul_expect; |
| 50 | |
| 51 | write_lock_bh(&nf_conntrack_lock); |
| 52 | nf_ct_unlink_expect(exp); |
| 53 | write_unlock_bh(&nf_conntrack_lock); |
| 54 | nf_conntrack_expect_put(exp); |
| 55 | } |
| 56 | |
| 57 | struct nf_conntrack_expect * |
| 58 | __nf_conntrack_expect_find(const struct nf_conntrack_tuple *tuple) |
| 59 | { |
| 60 | struct nf_conntrack_expect *i; |
| 61 | |
| 62 | list_for_each_entry(i, &nf_conntrack_expect_list, list) { |
| 63 | if (nf_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask)) |
| 64 | return i; |
| 65 | } |
| 66 | return NULL; |
| 67 | } |
| 68 | |
| 69 | /* Just find a expectation corresponding to a tuple. */ |
| 70 | struct nf_conntrack_expect * |
Yasuyuki Kozakai | 468ec44 | 2006-11-29 02:35:23 +0100 | [diff] [blame] | 71 | nf_conntrack_expect_find_get(const struct nf_conntrack_tuple *tuple) |
Martin Josefsson | 77ab9cf | 2006-11-29 02:34:58 +0100 | [diff] [blame] | 72 | { |
| 73 | struct nf_conntrack_expect *i; |
| 74 | |
| 75 | read_lock_bh(&nf_conntrack_lock); |
| 76 | i = __nf_conntrack_expect_find(tuple); |
| 77 | if (i) |
| 78 | atomic_inc(&i->use); |
| 79 | read_unlock_bh(&nf_conntrack_lock); |
| 80 | |
| 81 | return i; |
| 82 | } |
| 83 | |
| 84 | /* If an expectation for this connection is found, it gets delete from |
| 85 | * global list then returned. */ |
| 86 | struct nf_conntrack_expect * |
| 87 | find_expectation(const struct nf_conntrack_tuple *tuple) |
| 88 | { |
| 89 | struct nf_conntrack_expect *i; |
| 90 | |
| 91 | list_for_each_entry(i, &nf_conntrack_expect_list, list) { |
| 92 | /* If master is not in hash table yet (ie. packet hasn't left |
| 93 | this machine yet), how can other end know about expected? |
| 94 | Hence these are not the droids you are looking for (if |
| 95 | master ct never got confirmed, we'd hold a reference to it |
| 96 | and weird things would happen to future packets). */ |
| 97 | if (nf_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask) |
| 98 | && nf_ct_is_confirmed(i->master)) { |
| 99 | if (i->flags & NF_CT_EXPECT_PERMANENT) { |
| 100 | atomic_inc(&i->use); |
| 101 | return i; |
| 102 | } else if (del_timer(&i->timeout)) { |
| 103 | nf_ct_unlink_expect(i); |
| 104 | return i; |
| 105 | } |
| 106 | } |
| 107 | } |
| 108 | return NULL; |
| 109 | } |
| 110 | |
| 111 | /* delete all expectations for this conntrack */ |
| 112 | void nf_ct_remove_expectations(struct nf_conn *ct) |
| 113 | { |
| 114 | struct nf_conntrack_expect *i, *tmp; |
| 115 | struct nf_conn_help *help = nfct_help(ct); |
| 116 | |
| 117 | /* Optimization: most connection never expect any others. */ |
| 118 | if (!help || help->expecting == 0) |
| 119 | return; |
| 120 | |
| 121 | list_for_each_entry_safe(i, tmp, &nf_conntrack_expect_list, list) { |
| 122 | if (i->master == ct && del_timer(&i->timeout)) { |
| 123 | nf_ct_unlink_expect(i); |
| 124 | nf_conntrack_expect_put(i); |
| 125 | } |
| 126 | } |
| 127 | } |
| 128 | |
| 129 | /* Would two expected things clash? */ |
| 130 | static inline int expect_clash(const struct nf_conntrack_expect *a, |
| 131 | const struct nf_conntrack_expect *b) |
| 132 | { |
| 133 | /* Part covered by intersection of masks must be unequal, |
| 134 | otherwise they clash */ |
| 135 | struct nf_conntrack_tuple intersect_mask; |
| 136 | int count; |
| 137 | |
| 138 | intersect_mask.src.l3num = a->mask.src.l3num & b->mask.src.l3num; |
| 139 | intersect_mask.src.u.all = a->mask.src.u.all & b->mask.src.u.all; |
| 140 | intersect_mask.dst.u.all = a->mask.dst.u.all & b->mask.dst.u.all; |
| 141 | intersect_mask.dst.protonum = a->mask.dst.protonum |
| 142 | & b->mask.dst.protonum; |
| 143 | |
| 144 | for (count = 0; count < NF_CT_TUPLE_L3SIZE; count++){ |
| 145 | intersect_mask.src.u3.all[count] = |
| 146 | a->mask.src.u3.all[count] & b->mask.src.u3.all[count]; |
| 147 | } |
| 148 | |
| 149 | for (count = 0; count < NF_CT_TUPLE_L3SIZE; count++){ |
| 150 | intersect_mask.dst.u3.all[count] = |
| 151 | a->mask.dst.u3.all[count] & b->mask.dst.u3.all[count]; |
| 152 | } |
| 153 | |
| 154 | return nf_ct_tuple_mask_cmp(&a->tuple, &b->tuple, &intersect_mask); |
| 155 | } |
| 156 | |
| 157 | static inline int expect_matches(const struct nf_conntrack_expect *a, |
| 158 | const struct nf_conntrack_expect *b) |
| 159 | { |
| 160 | return a->master == b->master |
| 161 | && nf_ct_tuple_equal(&a->tuple, &b->tuple) |
| 162 | && nf_ct_tuple_equal(&a->mask, &b->mask); |
| 163 | } |
| 164 | |
| 165 | /* Generally a bad idea to call this: could have matched already. */ |
| 166 | void nf_conntrack_unexpect_related(struct nf_conntrack_expect *exp) |
| 167 | { |
| 168 | struct nf_conntrack_expect *i; |
| 169 | |
| 170 | write_lock_bh(&nf_conntrack_lock); |
| 171 | /* choose the the oldest expectation to evict */ |
| 172 | list_for_each_entry_reverse(i, &nf_conntrack_expect_list, list) { |
| 173 | if (expect_matches(i, exp) && del_timer(&i->timeout)) { |
| 174 | nf_ct_unlink_expect(i); |
| 175 | write_unlock_bh(&nf_conntrack_lock); |
| 176 | nf_conntrack_expect_put(i); |
| 177 | return; |
| 178 | } |
| 179 | } |
| 180 | write_unlock_bh(&nf_conntrack_lock); |
| 181 | } |
| 182 | |
| 183 | /* We don't increase the master conntrack refcount for non-fulfilled |
| 184 | * conntracks. During the conntrack destruction, the expectations are |
| 185 | * always killed before the conntrack itself */ |
| 186 | struct nf_conntrack_expect *nf_conntrack_expect_alloc(struct nf_conn *me) |
| 187 | { |
| 188 | struct nf_conntrack_expect *new; |
| 189 | |
| 190 | new = kmem_cache_alloc(nf_conntrack_expect_cachep, GFP_ATOMIC); |
| 191 | if (!new) |
| 192 | return NULL; |
| 193 | |
| 194 | new->master = me; |
| 195 | atomic_set(&new->use, 1); |
| 196 | return new; |
| 197 | } |
| 198 | |
Patrick McHardy | d6a9b65 | 2006-12-02 22:08:01 -0800 | [diff] [blame^] | 199 | void nf_conntrack_expect_init(struct nf_conntrack_expect *exp, int family, |
| 200 | union nf_conntrack_address *saddr, |
| 201 | union nf_conntrack_address *daddr, |
| 202 | u_int8_t proto, __be16 *src, __be16 *dst) |
| 203 | { |
| 204 | int len; |
| 205 | |
| 206 | if (family == AF_INET) |
| 207 | len = 4; |
| 208 | else |
| 209 | len = 16; |
| 210 | |
| 211 | exp->flags = 0; |
| 212 | exp->expectfn = NULL; |
| 213 | exp->helper = NULL; |
| 214 | exp->tuple.src.l3num = family; |
| 215 | exp->tuple.dst.protonum = proto; |
| 216 | exp->mask.src.l3num = 0xFFFF; |
| 217 | exp->mask.dst.protonum = 0xFF; |
| 218 | |
| 219 | if (saddr) { |
| 220 | memcpy(&exp->tuple.src.u3, saddr, len); |
| 221 | if (sizeof(exp->tuple.src.u3) > len) |
| 222 | /* address needs to be cleared for nf_ct_tuple_equal */ |
| 223 | memset((void *)&exp->tuple.src.u3 + len, 0x00, |
| 224 | sizeof(exp->tuple.src.u3) - len); |
| 225 | memset(&exp->mask.src.u3, 0xFF, len); |
| 226 | if (sizeof(exp->mask.src.u3) > len) |
| 227 | memset((void *)&exp->mask.src.u3 + len, 0x00, |
| 228 | sizeof(exp->mask.src.u3) - len); |
| 229 | } else { |
| 230 | memset(&exp->tuple.src.u3, 0x00, sizeof(exp->tuple.src.u3)); |
| 231 | memset(&exp->mask.src.u3, 0x00, sizeof(exp->mask.src.u3)); |
| 232 | } |
| 233 | |
| 234 | if (daddr) { |
| 235 | memcpy(&exp->tuple.dst.u3, daddr, len); |
| 236 | if (sizeof(exp->tuple.dst.u3) > len) |
| 237 | /* address needs to be cleared for nf_ct_tuple_equal */ |
| 238 | memset((void *)&exp->tuple.dst.u3 + len, 0x00, |
| 239 | sizeof(exp->tuple.dst.u3) - len); |
| 240 | memset(&exp->mask.dst.u3, 0xFF, len); |
| 241 | if (sizeof(exp->mask.dst.u3) > len) |
| 242 | memset((void *)&exp->mask.dst.u3 + len, 0x00, |
| 243 | sizeof(exp->mask.dst.u3) - len); |
| 244 | } else { |
| 245 | memset(&exp->tuple.dst.u3, 0x00, sizeof(exp->tuple.dst.u3)); |
| 246 | memset(&exp->mask.dst.u3, 0x00, sizeof(exp->mask.dst.u3)); |
| 247 | } |
| 248 | |
| 249 | if (src) { |
| 250 | exp->tuple.src.u.all = (__force u16)*src; |
| 251 | exp->mask.src.u.all = 0xFFFF; |
| 252 | } else { |
| 253 | exp->tuple.src.u.all = 0; |
| 254 | exp->mask.src.u.all = 0; |
| 255 | } |
| 256 | |
| 257 | if (dst) { |
| 258 | exp->tuple.dst.u.all = (__force u16)*dst; |
| 259 | exp->mask.dst.u.all = 0xFFFF; |
| 260 | } else { |
| 261 | exp->tuple.dst.u.all = 0; |
| 262 | exp->mask.dst.u.all = 0; |
| 263 | } |
| 264 | } |
| 265 | EXPORT_SYMBOL_GPL(nf_conntrack_expect_init); |
| 266 | |
Martin Josefsson | 77ab9cf | 2006-11-29 02:34:58 +0100 | [diff] [blame] | 267 | void nf_conntrack_expect_put(struct nf_conntrack_expect *exp) |
| 268 | { |
| 269 | if (atomic_dec_and_test(&exp->use)) |
| 270 | kmem_cache_free(nf_conntrack_expect_cachep, exp); |
| 271 | } |
| 272 | |
| 273 | static void nf_conntrack_expect_insert(struct nf_conntrack_expect *exp) |
| 274 | { |
| 275 | struct nf_conn_help *master_help = nfct_help(exp->master); |
| 276 | |
| 277 | atomic_inc(&exp->use); |
| 278 | master_help->expecting++; |
| 279 | list_add(&exp->list, &nf_conntrack_expect_list); |
| 280 | |
| 281 | init_timer(&exp->timeout); |
| 282 | exp->timeout.data = (unsigned long)exp; |
| 283 | exp->timeout.function = expectation_timed_out; |
| 284 | exp->timeout.expires = jiffies + master_help->helper->timeout * HZ; |
| 285 | add_timer(&exp->timeout); |
| 286 | |
| 287 | exp->id = ++nf_conntrack_expect_next_id; |
| 288 | atomic_inc(&exp->use); |
| 289 | NF_CT_STAT_INC(expect_create); |
| 290 | } |
| 291 | |
| 292 | /* Race with expectations being used means we could have none to find; OK. */ |
| 293 | static void evict_oldest_expect(struct nf_conn *master) |
| 294 | { |
| 295 | struct nf_conntrack_expect *i; |
| 296 | |
| 297 | list_for_each_entry_reverse(i, &nf_conntrack_expect_list, list) { |
| 298 | if (i->master == master) { |
| 299 | if (del_timer(&i->timeout)) { |
| 300 | nf_ct_unlink_expect(i); |
| 301 | nf_conntrack_expect_put(i); |
| 302 | } |
| 303 | break; |
| 304 | } |
| 305 | } |
| 306 | } |
| 307 | |
| 308 | static inline int refresh_timer(struct nf_conntrack_expect *i) |
| 309 | { |
| 310 | struct nf_conn_help *master_help = nfct_help(i->master); |
| 311 | |
| 312 | if (!del_timer(&i->timeout)) |
| 313 | return 0; |
| 314 | |
| 315 | i->timeout.expires = jiffies + master_help->helper->timeout*HZ; |
| 316 | add_timer(&i->timeout); |
| 317 | return 1; |
| 318 | } |
| 319 | |
| 320 | int nf_conntrack_expect_related(struct nf_conntrack_expect *expect) |
| 321 | { |
| 322 | struct nf_conntrack_expect *i; |
| 323 | struct nf_conn *master = expect->master; |
| 324 | struct nf_conn_help *master_help = nfct_help(master); |
| 325 | int ret; |
| 326 | |
| 327 | NF_CT_ASSERT(master_help); |
| 328 | |
| 329 | write_lock_bh(&nf_conntrack_lock); |
| 330 | list_for_each_entry(i, &nf_conntrack_expect_list, list) { |
| 331 | if (expect_matches(i, expect)) { |
| 332 | /* Refresh timer: if it's dying, ignore.. */ |
| 333 | if (refresh_timer(i)) { |
| 334 | ret = 0; |
| 335 | goto out; |
| 336 | } |
| 337 | } else if (expect_clash(i, expect)) { |
| 338 | ret = -EBUSY; |
| 339 | goto out; |
| 340 | } |
| 341 | } |
| 342 | /* Will be over limit? */ |
| 343 | if (master_help->helper->max_expected && |
| 344 | master_help->expecting >= master_help->helper->max_expected) |
| 345 | evict_oldest_expect(master); |
| 346 | |
| 347 | nf_conntrack_expect_insert(expect); |
| 348 | nf_conntrack_expect_event(IPEXP_NEW, expect); |
| 349 | ret = 0; |
| 350 | out: |
| 351 | write_unlock_bh(&nf_conntrack_lock); |
| 352 | return ret; |
| 353 | } |
| 354 | |
| 355 | #ifdef CONFIG_PROC_FS |
| 356 | static void *exp_seq_start(struct seq_file *s, loff_t *pos) |
| 357 | { |
| 358 | struct list_head *e = &nf_conntrack_expect_list; |
| 359 | loff_t i; |
| 360 | |
| 361 | /* strange seq_file api calls stop even if we fail, |
| 362 | * thus we need to grab lock since stop unlocks */ |
| 363 | read_lock_bh(&nf_conntrack_lock); |
| 364 | |
| 365 | if (list_empty(e)) |
| 366 | return NULL; |
| 367 | |
| 368 | for (i = 0; i <= *pos; i++) { |
| 369 | e = e->next; |
| 370 | if (e == &nf_conntrack_expect_list) |
| 371 | return NULL; |
| 372 | } |
| 373 | return e; |
| 374 | } |
| 375 | |
| 376 | static void *exp_seq_next(struct seq_file *s, void *v, loff_t *pos) |
| 377 | { |
| 378 | struct list_head *e = v; |
| 379 | |
| 380 | ++*pos; |
| 381 | e = e->next; |
| 382 | |
| 383 | if (e == &nf_conntrack_expect_list) |
| 384 | return NULL; |
| 385 | |
| 386 | return e; |
| 387 | } |
| 388 | |
| 389 | static void exp_seq_stop(struct seq_file *s, void *v) |
| 390 | { |
| 391 | read_unlock_bh(&nf_conntrack_lock); |
| 392 | } |
| 393 | |
| 394 | static int exp_seq_show(struct seq_file *s, void *v) |
| 395 | { |
| 396 | struct nf_conntrack_expect *expect = v; |
| 397 | |
| 398 | if (expect->timeout.function) |
| 399 | seq_printf(s, "%ld ", timer_pending(&expect->timeout) |
| 400 | ? (long)(expect->timeout.expires - jiffies)/HZ : 0); |
| 401 | else |
| 402 | seq_printf(s, "- "); |
| 403 | seq_printf(s, "l3proto = %u proto=%u ", |
| 404 | expect->tuple.src.l3num, |
| 405 | expect->tuple.dst.protonum); |
| 406 | print_tuple(s, &expect->tuple, |
| 407 | __nf_ct_l3proto_find(expect->tuple.src.l3num), |
Martin Josefsson | 605dcad | 2006-11-29 02:35:06 +0100 | [diff] [blame] | 408 | __nf_ct_l4proto_find(expect->tuple.src.l3num, |
Martin Josefsson | 77ab9cf | 2006-11-29 02:34:58 +0100 | [diff] [blame] | 409 | expect->tuple.dst.protonum)); |
| 410 | return seq_putc(s, '\n'); |
| 411 | } |
| 412 | |
| 413 | static struct seq_operations exp_seq_ops = { |
| 414 | .start = exp_seq_start, |
| 415 | .next = exp_seq_next, |
| 416 | .stop = exp_seq_stop, |
| 417 | .show = exp_seq_show |
| 418 | }; |
| 419 | |
| 420 | static int exp_open(struct inode *inode, struct file *file) |
| 421 | { |
| 422 | return seq_open(file, &exp_seq_ops); |
| 423 | } |
| 424 | |
| 425 | struct file_operations exp_file_ops = { |
| 426 | .owner = THIS_MODULE, |
| 427 | .open = exp_open, |
| 428 | .read = seq_read, |
| 429 | .llseek = seq_lseek, |
| 430 | .release = seq_release |
| 431 | }; |
| 432 | #endif /* CONFIG_PROC_FS */ |