Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* iptables match extension to limit the number of packets per second |
| 2 | * seperately for each hashbucket (sourceip/sourceport/dstip/dstport) |
| 3 | * |
| 4 | * (C) 2003-2004 by Harald Welte <laforge@netfilter.org> |
| 5 | * |
| 6 | * $Id: ipt_hashlimit.c 3244 2004-10-20 16:24:29Z laforge@netfilter.org $ |
| 7 | * |
| 8 | * Development of this code was funded by Astaro AG, http://www.astaro.com/ |
| 9 | * |
| 10 | * based on ipt_limit.c by: |
| 11 | * Jérôme de Vivie <devivie@info.enserb.u-bordeaux.fr> |
| 12 | * Hervé Eychenne <eychenne@info.enserb.u-bordeaux.fr> |
| 13 | * Rusty Russell <rusty@rustcorp.com.au> |
| 14 | * |
| 15 | * The general idea is to create a hash table for every dstip and have a |
| 16 | * seperate limit counter per tuple. This way you can do something like 'limit |
| 17 | * the number of syn packets for each of my internal addresses. |
| 18 | * |
| 19 | * Ideally this would just be implemented as a general 'hash' match, which would |
| 20 | * allow us to attach any iptables target to it's hash buckets. But this is |
| 21 | * not possible in the current iptables architecture. As always, pkttables for |
| 22 | * 2.7.x will help ;) |
| 23 | */ |
| 24 | #include <linux/module.h> |
| 25 | #include <linux/skbuff.h> |
| 26 | #include <linux/spinlock.h> |
| 27 | #include <linux/random.h> |
| 28 | #include <linux/jhash.h> |
| 29 | #include <linux/slab.h> |
| 30 | #include <linux/vmalloc.h> |
| 31 | #include <linux/tcp.h> |
| 32 | #include <linux/udp.h> |
| 33 | #include <linux/sctp.h> |
| 34 | #include <linux/proc_fs.h> |
| 35 | #include <linux/seq_file.h> |
| 36 | #include <linux/list.h> |
| 37 | |
| 38 | #include <linux/netfilter_ipv4/ip_tables.h> |
| 39 | #include <linux/netfilter_ipv4/ipt_hashlimit.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 40 | |
| 41 | /* FIXME: this is just for IP_NF_ASSERRT */ |
| 42 | #include <linux/netfilter_ipv4/ip_conntrack.h> |
| 43 | |
| 44 | MODULE_LICENSE("GPL"); |
| 45 | MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>"); |
| 46 | MODULE_DESCRIPTION("iptables match for limiting per hash-bucket"); |
| 47 | |
| 48 | /* need to declare this at the top */ |
| 49 | static struct proc_dir_entry *hashlimit_procdir; |
| 50 | static struct file_operations dl_file_ops; |
| 51 | |
| 52 | /* hash table crap */ |
| 53 | |
| 54 | struct dsthash_dst { |
| 55 | u_int32_t src_ip; |
| 56 | u_int32_t dst_ip; |
| 57 | /* ports have to be consecutive !!! */ |
| 58 | u_int16_t src_port; |
| 59 | u_int16_t dst_port; |
| 60 | }; |
| 61 | |
| 62 | struct dsthash_ent { |
| 63 | /* static / read-only parts in the beginning */ |
| 64 | struct hlist_node node; |
| 65 | struct dsthash_dst dst; |
| 66 | |
| 67 | /* modified structure members in the end */ |
| 68 | unsigned long expires; /* precalculated expiry time */ |
| 69 | struct { |
| 70 | unsigned long prev; /* last modification */ |
| 71 | u_int32_t credit; |
| 72 | u_int32_t credit_cap, cost; |
| 73 | } rateinfo; |
| 74 | }; |
| 75 | |
| 76 | struct ipt_hashlimit_htable { |
| 77 | struct hlist_node node; /* global list of all htables */ |
| 78 | atomic_t use; |
| 79 | |
| 80 | struct hashlimit_cfg cfg; /* config */ |
| 81 | |
| 82 | /* used internally */ |
| 83 | spinlock_t lock; /* lock for list_head */ |
| 84 | u_int32_t rnd; /* random seed for hash */ |
| 85 | struct timer_list timer; /* timer for gc */ |
| 86 | atomic_t count; /* number entries in table */ |
| 87 | |
| 88 | /* seq_file stuff */ |
| 89 | struct proc_dir_entry *pde; |
| 90 | |
| 91 | struct hlist_head hash[0]; /* hashtable itself */ |
| 92 | }; |
| 93 | |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 94 | static DEFINE_SPINLOCK(hashlimit_lock); /* protects htables list */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 95 | static DECLARE_MUTEX(hlimit_mutex); /* additional checkentry protection */ |
| 96 | static HLIST_HEAD(hashlimit_htables); |
Eric Dumazet | ba89966 | 2005-08-26 12:05:31 -0700 | [diff] [blame] | 97 | static kmem_cache_t *hashlimit_cachep __read_mostly; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 98 | |
| 99 | static inline int dst_cmp(const struct dsthash_ent *ent, struct dsthash_dst *b) |
| 100 | { |
| 101 | return (ent->dst.dst_ip == b->dst_ip |
| 102 | && ent->dst.dst_port == b->dst_port |
| 103 | && ent->dst.src_port == b->src_port |
| 104 | && ent->dst.src_ip == b->src_ip); |
| 105 | } |
| 106 | |
| 107 | static inline u_int32_t |
| 108 | hash_dst(const struct ipt_hashlimit_htable *ht, const struct dsthash_dst *dst) |
| 109 | { |
| 110 | return (jhash_3words(dst->dst_ip, (dst->dst_port<<16 | dst->src_port), |
| 111 | dst->src_ip, ht->rnd) % ht->cfg.size); |
| 112 | } |
| 113 | |
| 114 | static inline struct dsthash_ent * |
| 115 | __dsthash_find(const struct ipt_hashlimit_htable *ht, struct dsthash_dst *dst) |
| 116 | { |
| 117 | struct dsthash_ent *ent; |
| 118 | struct hlist_node *pos; |
| 119 | u_int32_t hash = hash_dst(ht, dst); |
| 120 | |
| 121 | if (!hlist_empty(&ht->hash[hash])) |
| 122 | hlist_for_each_entry(ent, pos, &ht->hash[hash], node) { |
| 123 | if (dst_cmp(ent, dst)) { |
| 124 | return ent; |
| 125 | } |
| 126 | } |
| 127 | |
| 128 | return NULL; |
| 129 | } |
| 130 | |
| 131 | /* allocate dsthash_ent, initialize dst, put in htable and lock it */ |
| 132 | static struct dsthash_ent * |
| 133 | __dsthash_alloc_init(struct ipt_hashlimit_htable *ht, struct dsthash_dst *dst) |
| 134 | { |
| 135 | struct dsthash_ent *ent; |
| 136 | |
| 137 | /* initialize hash with random val at the time we allocate |
| 138 | * the first hashtable entry */ |
| 139 | if (!ht->rnd) |
| 140 | get_random_bytes(&ht->rnd, 4); |
| 141 | |
| 142 | if (ht->cfg.max && |
| 143 | atomic_read(&ht->count) >= ht->cfg.max) { |
| 144 | /* FIXME: do something. question is what.. */ |
| 145 | if (net_ratelimit()) |
| 146 | printk(KERN_WARNING |
| 147 | "ipt_hashlimit: max count of %u reached\n", |
| 148 | ht->cfg.max); |
| 149 | return NULL; |
| 150 | } |
| 151 | |
| 152 | ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC); |
| 153 | if (!ent) { |
| 154 | if (net_ratelimit()) |
| 155 | printk(KERN_ERR |
| 156 | "ipt_hashlimit: can't allocate dsthash_ent\n"); |
| 157 | return NULL; |
| 158 | } |
| 159 | |
| 160 | atomic_inc(&ht->count); |
| 161 | |
| 162 | ent->dst.dst_ip = dst->dst_ip; |
| 163 | ent->dst.dst_port = dst->dst_port; |
| 164 | ent->dst.src_ip = dst->src_ip; |
| 165 | ent->dst.src_port = dst->src_port; |
| 166 | |
| 167 | hlist_add_head(&ent->node, &ht->hash[hash_dst(ht, dst)]); |
| 168 | |
| 169 | return ent; |
| 170 | } |
| 171 | |
| 172 | static inline void |
| 173 | __dsthash_free(struct ipt_hashlimit_htable *ht, struct dsthash_ent *ent) |
| 174 | { |
| 175 | hlist_del(&ent->node); |
| 176 | kmem_cache_free(hashlimit_cachep, ent); |
| 177 | atomic_dec(&ht->count); |
| 178 | } |
| 179 | static void htable_gc(unsigned long htlong); |
| 180 | |
| 181 | static int htable_create(struct ipt_hashlimit_info *minfo) |
| 182 | { |
| 183 | int i; |
| 184 | unsigned int size; |
| 185 | struct ipt_hashlimit_htable *hinfo; |
| 186 | |
| 187 | if (minfo->cfg.size) |
| 188 | size = minfo->cfg.size; |
| 189 | else { |
| 190 | size = (((num_physpages << PAGE_SHIFT) / 16384) |
| 191 | / sizeof(struct list_head)); |
| 192 | if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE)) |
| 193 | size = 8192; |
| 194 | if (size < 16) |
| 195 | size = 16; |
| 196 | } |
| 197 | /* FIXME: don't use vmalloc() here or anywhere else -HW */ |
| 198 | hinfo = vmalloc(sizeof(struct ipt_hashlimit_htable) |
| 199 | + (sizeof(struct list_head) * size)); |
| 200 | if (!hinfo) { |
| 201 | printk(KERN_ERR "ipt_hashlimit: Unable to create hashtable\n"); |
| 202 | return -1; |
| 203 | } |
| 204 | minfo->hinfo = hinfo; |
| 205 | |
| 206 | /* copy match config into hashtable config */ |
| 207 | memcpy(&hinfo->cfg, &minfo->cfg, sizeof(hinfo->cfg)); |
| 208 | hinfo->cfg.size = size; |
| 209 | if (!hinfo->cfg.max) |
| 210 | hinfo->cfg.max = 8 * hinfo->cfg.size; |
| 211 | else if (hinfo->cfg.max < hinfo->cfg.size) |
| 212 | hinfo->cfg.max = hinfo->cfg.size; |
| 213 | |
| 214 | for (i = 0; i < hinfo->cfg.size; i++) |
| 215 | INIT_HLIST_HEAD(&hinfo->hash[i]); |
| 216 | |
| 217 | atomic_set(&hinfo->count, 0); |
| 218 | atomic_set(&hinfo->use, 1); |
| 219 | hinfo->rnd = 0; |
| 220 | spin_lock_init(&hinfo->lock); |
| 221 | hinfo->pde = create_proc_entry(minfo->name, 0, hashlimit_procdir); |
| 222 | if (!hinfo->pde) { |
| 223 | vfree(hinfo); |
| 224 | return -1; |
| 225 | } |
| 226 | hinfo->pde->proc_fops = &dl_file_ops; |
| 227 | hinfo->pde->data = hinfo; |
| 228 | |
| 229 | init_timer(&hinfo->timer); |
| 230 | hinfo->timer.expires = jiffies + msecs_to_jiffies(hinfo->cfg.gc_interval); |
| 231 | hinfo->timer.data = (unsigned long )hinfo; |
| 232 | hinfo->timer.function = htable_gc; |
| 233 | add_timer(&hinfo->timer); |
| 234 | |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 235 | spin_lock_bh(&hashlimit_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 236 | hlist_add_head(&hinfo->node, &hashlimit_htables); |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 237 | spin_unlock_bh(&hashlimit_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 238 | |
| 239 | return 0; |
| 240 | } |
| 241 | |
| 242 | static int select_all(struct ipt_hashlimit_htable *ht, struct dsthash_ent *he) |
| 243 | { |
| 244 | return 1; |
| 245 | } |
| 246 | |
| 247 | static int select_gc(struct ipt_hashlimit_htable *ht, struct dsthash_ent *he) |
| 248 | { |
| 249 | return (jiffies >= he->expires); |
| 250 | } |
| 251 | |
| 252 | static void htable_selective_cleanup(struct ipt_hashlimit_htable *ht, |
| 253 | int (*select)(struct ipt_hashlimit_htable *ht, |
| 254 | struct dsthash_ent *he)) |
| 255 | { |
| 256 | int i; |
| 257 | |
| 258 | IP_NF_ASSERT(ht->cfg.size && ht->cfg.max); |
| 259 | |
| 260 | /* lock hash table and iterate over it */ |
| 261 | spin_lock_bh(&ht->lock); |
| 262 | for (i = 0; i < ht->cfg.size; i++) { |
| 263 | struct dsthash_ent *dh; |
| 264 | struct hlist_node *pos, *n; |
| 265 | hlist_for_each_entry_safe(dh, pos, n, &ht->hash[i], node) { |
| 266 | if ((*select)(ht, dh)) |
| 267 | __dsthash_free(ht, dh); |
| 268 | } |
| 269 | } |
| 270 | spin_unlock_bh(&ht->lock); |
| 271 | } |
| 272 | |
| 273 | /* hash table garbage collector, run by timer */ |
| 274 | static void htable_gc(unsigned long htlong) |
| 275 | { |
| 276 | struct ipt_hashlimit_htable *ht = (struct ipt_hashlimit_htable *)htlong; |
| 277 | |
| 278 | htable_selective_cleanup(ht, select_gc); |
| 279 | |
| 280 | /* re-add the timer accordingly */ |
| 281 | ht->timer.expires = jiffies + msecs_to_jiffies(ht->cfg.gc_interval); |
| 282 | add_timer(&ht->timer); |
| 283 | } |
| 284 | |
| 285 | static void htable_destroy(struct ipt_hashlimit_htable *hinfo) |
| 286 | { |
| 287 | /* remove timer, if it is pending */ |
| 288 | if (timer_pending(&hinfo->timer)) |
| 289 | del_timer(&hinfo->timer); |
| 290 | |
| 291 | /* remove proc entry */ |
| 292 | remove_proc_entry(hinfo->pde->name, hashlimit_procdir); |
| 293 | |
| 294 | htable_selective_cleanup(hinfo, select_all); |
| 295 | vfree(hinfo); |
| 296 | } |
| 297 | |
| 298 | static struct ipt_hashlimit_htable *htable_find_get(char *name) |
| 299 | { |
| 300 | struct ipt_hashlimit_htable *hinfo; |
| 301 | struct hlist_node *pos; |
| 302 | |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 303 | spin_lock_bh(&hashlimit_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 304 | hlist_for_each_entry(hinfo, pos, &hashlimit_htables, node) { |
| 305 | if (!strcmp(name, hinfo->pde->name)) { |
| 306 | atomic_inc(&hinfo->use); |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 307 | spin_unlock_bh(&hashlimit_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 308 | return hinfo; |
| 309 | } |
| 310 | } |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 311 | spin_unlock_bh(&hashlimit_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 312 | |
| 313 | return NULL; |
| 314 | } |
| 315 | |
| 316 | static void htable_put(struct ipt_hashlimit_htable *hinfo) |
| 317 | { |
| 318 | if (atomic_dec_and_test(&hinfo->use)) { |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 319 | spin_lock_bh(&hashlimit_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 320 | hlist_del(&hinfo->node); |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 321 | spin_unlock_bh(&hashlimit_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 322 | htable_destroy(hinfo); |
| 323 | } |
| 324 | } |
| 325 | |
| 326 | |
| 327 | /* The algorithm used is the Simple Token Bucket Filter (TBF) |
| 328 | * see net/sched/sch_tbf.c in the linux source tree |
| 329 | */ |
| 330 | |
| 331 | /* Rusty: This is my (non-mathematically-inclined) understanding of |
| 332 | this algorithm. The `average rate' in jiffies becomes your initial |
| 333 | amount of credit `credit' and the most credit you can ever have |
| 334 | `credit_cap'. The `peak rate' becomes the cost of passing the |
| 335 | test, `cost'. |
| 336 | |
| 337 | `prev' tracks the last packet hit: you gain one credit per jiffy. |
| 338 | If you get credit balance more than this, the extra credit is |
| 339 | discarded. Every time the match passes, you lose `cost' credits; |
| 340 | if you don't have that many, the test fails. |
| 341 | |
| 342 | See Alexey's formal explanation in net/sched/sch_tbf.c. |
| 343 | |
| 344 | To get the maximum range, we multiply by this factor (ie. you get N |
| 345 | credits per jiffy). We want to allow a rate as low as 1 per day |
| 346 | (slowest userspace tool allows), which means |
| 347 | CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie. |
| 348 | */ |
| 349 | #define MAX_CPJ (0xFFFFFFFF / (HZ*60*60*24)) |
| 350 | |
| 351 | /* Repeated shift and or gives us all 1s, final shift and add 1 gives |
| 352 | * us the power of 2 below the theoretical max, so GCC simply does a |
| 353 | * shift. */ |
| 354 | #define _POW2_BELOW2(x) ((x)|((x)>>1)) |
| 355 | #define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2)) |
| 356 | #define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4)) |
| 357 | #define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8)) |
| 358 | #define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16)) |
| 359 | #define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1) |
| 360 | |
| 361 | #define CREDITS_PER_JIFFY POW2_BELOW32(MAX_CPJ) |
| 362 | |
| 363 | /* Precision saver. */ |
| 364 | static inline u_int32_t |
| 365 | user2credits(u_int32_t user) |
| 366 | { |
| 367 | /* If multiplying would overflow... */ |
| 368 | if (user > 0xFFFFFFFF / (HZ*CREDITS_PER_JIFFY)) |
| 369 | /* Divide first. */ |
| 370 | return (user / IPT_HASHLIMIT_SCALE) * HZ * CREDITS_PER_JIFFY; |
| 371 | |
| 372 | return (user * HZ * CREDITS_PER_JIFFY) / IPT_HASHLIMIT_SCALE; |
| 373 | } |
| 374 | |
| 375 | static inline void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now) |
| 376 | { |
| 377 | dh->rateinfo.credit += (now - xchg(&dh->rateinfo.prev, now)) |
| 378 | * CREDITS_PER_JIFFY; |
| 379 | if (dh->rateinfo.credit > dh->rateinfo.credit_cap) |
| 380 | dh->rateinfo.credit = dh->rateinfo.credit_cap; |
| 381 | } |
| 382 | |
| 383 | static inline int get_ports(const struct sk_buff *skb, int offset, |
| 384 | u16 ports[2]) |
| 385 | { |
| 386 | union { |
| 387 | struct tcphdr th; |
| 388 | struct udphdr uh; |
| 389 | sctp_sctphdr_t sctph; |
| 390 | } hdr_u, *ptr_u; |
| 391 | |
| 392 | /* Must not be a fragment. */ |
| 393 | if (offset) |
| 394 | return 1; |
| 395 | |
| 396 | /* Must be big enough to read ports (both UDP and TCP have |
| 397 | them at the start). */ |
| 398 | ptr_u = skb_header_pointer(skb, skb->nh.iph->ihl*4, 8, &hdr_u); |
| 399 | if (!ptr_u) |
| 400 | return 1; |
| 401 | |
| 402 | switch (skb->nh.iph->protocol) { |
| 403 | case IPPROTO_TCP: |
| 404 | ports[0] = ptr_u->th.source; |
| 405 | ports[1] = ptr_u->th.dest; |
| 406 | break; |
| 407 | case IPPROTO_UDP: |
| 408 | ports[0] = ptr_u->uh.source; |
| 409 | ports[1] = ptr_u->uh.dest; |
| 410 | break; |
| 411 | case IPPROTO_SCTP: |
| 412 | ports[0] = ptr_u->sctph.source; |
| 413 | ports[1] = ptr_u->sctph.dest; |
| 414 | break; |
| 415 | default: |
| 416 | /* all other protocols don't supprot per-port hash |
| 417 | * buckets */ |
| 418 | ports[0] = ports[1] = 0; |
| 419 | break; |
| 420 | } |
| 421 | |
| 422 | return 0; |
| 423 | } |
| 424 | |
| 425 | |
| 426 | static int |
| 427 | hashlimit_match(const struct sk_buff *skb, |
| 428 | const struct net_device *in, |
| 429 | const struct net_device *out, |
| 430 | const void *matchinfo, |
| 431 | int offset, |
Harald Welte | 2e4e6a1 | 2006-01-12 13:30:04 -0800 | [diff] [blame] | 432 | unsigned int protoff, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 433 | int *hotdrop) |
| 434 | { |
| 435 | struct ipt_hashlimit_info *r = |
| 436 | ((struct ipt_hashlimit_info *)matchinfo)->u.master; |
| 437 | struct ipt_hashlimit_htable *hinfo = r->hinfo; |
| 438 | unsigned long now = jiffies; |
| 439 | struct dsthash_ent *dh; |
| 440 | struct dsthash_dst dst; |
| 441 | |
| 442 | /* build 'dst' according to hinfo->cfg and current packet */ |
| 443 | memset(&dst, 0, sizeof(dst)); |
| 444 | if (hinfo->cfg.mode & IPT_HASHLIMIT_HASH_DIP) |
| 445 | dst.dst_ip = skb->nh.iph->daddr; |
| 446 | if (hinfo->cfg.mode & IPT_HASHLIMIT_HASH_SIP) |
| 447 | dst.src_ip = skb->nh.iph->saddr; |
| 448 | if (hinfo->cfg.mode & IPT_HASHLIMIT_HASH_DPT |
| 449 | ||hinfo->cfg.mode & IPT_HASHLIMIT_HASH_SPT) { |
| 450 | u_int16_t ports[2]; |
| 451 | if (get_ports(skb, offset, ports)) { |
| 452 | /* We've been asked to examine this packet, and we |
| 453 | can't. Hence, no choice but to drop. */ |
| 454 | *hotdrop = 1; |
| 455 | return 0; |
| 456 | } |
| 457 | if (hinfo->cfg.mode & IPT_HASHLIMIT_HASH_SPT) |
| 458 | dst.src_port = ports[0]; |
| 459 | if (hinfo->cfg.mode & IPT_HASHLIMIT_HASH_DPT) |
| 460 | dst.dst_port = ports[1]; |
| 461 | } |
| 462 | |
| 463 | spin_lock_bh(&hinfo->lock); |
| 464 | dh = __dsthash_find(hinfo, &dst); |
| 465 | if (!dh) { |
| 466 | dh = __dsthash_alloc_init(hinfo, &dst); |
| 467 | |
| 468 | if (!dh) { |
| 469 | /* enomem... don't match == DROP */ |
| 470 | if (net_ratelimit()) |
| 471 | printk(KERN_ERR "%s: ENOMEM\n", __FUNCTION__); |
| 472 | spin_unlock_bh(&hinfo->lock); |
| 473 | return 0; |
| 474 | } |
| 475 | |
| 476 | dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire); |
| 477 | |
| 478 | dh->rateinfo.prev = jiffies; |
| 479 | dh->rateinfo.credit = user2credits(hinfo->cfg.avg * |
| 480 | hinfo->cfg.burst); |
| 481 | dh->rateinfo.credit_cap = user2credits(hinfo->cfg.avg * |
| 482 | hinfo->cfg.burst); |
| 483 | dh->rateinfo.cost = user2credits(hinfo->cfg.avg); |
| 484 | |
| 485 | spin_unlock_bh(&hinfo->lock); |
| 486 | return 1; |
| 487 | } |
| 488 | |
| 489 | /* update expiration timeout */ |
| 490 | dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire); |
| 491 | |
| 492 | rateinfo_recalc(dh, now); |
| 493 | if (dh->rateinfo.credit >= dh->rateinfo.cost) { |
| 494 | /* We're underlimit. */ |
| 495 | dh->rateinfo.credit -= dh->rateinfo.cost; |
| 496 | spin_unlock_bh(&hinfo->lock); |
| 497 | return 1; |
| 498 | } |
| 499 | |
| 500 | spin_unlock_bh(&hinfo->lock); |
| 501 | |
| 502 | /* default case: we're overlimit, thus don't match */ |
| 503 | return 0; |
| 504 | } |
| 505 | |
| 506 | static int |
| 507 | hashlimit_checkentry(const char *tablename, |
Harald Welte | 2e4e6a1 | 2006-01-12 13:30:04 -0800 | [diff] [blame] | 508 | const void *inf, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 509 | void *matchinfo, |
| 510 | unsigned int matchsize, |
| 511 | unsigned int hook_mask) |
| 512 | { |
| 513 | struct ipt_hashlimit_info *r = matchinfo; |
| 514 | |
| 515 | if (matchsize != IPT_ALIGN(sizeof(struct ipt_hashlimit_info))) |
| 516 | return 0; |
| 517 | |
| 518 | /* Check for overflow. */ |
| 519 | if (r->cfg.burst == 0 |
| 520 | || user2credits(r->cfg.avg * r->cfg.burst) < |
| 521 | user2credits(r->cfg.avg)) { |
| 522 | printk(KERN_ERR "ipt_hashlimit: Overflow, try lower: %u/%u\n", |
| 523 | r->cfg.avg, r->cfg.burst); |
| 524 | return 0; |
| 525 | } |
| 526 | |
| 527 | if (r->cfg.mode == 0 |
| 528 | || r->cfg.mode > (IPT_HASHLIMIT_HASH_DPT |
| 529 | |IPT_HASHLIMIT_HASH_DIP |
| 530 | |IPT_HASHLIMIT_HASH_SIP |
| 531 | |IPT_HASHLIMIT_HASH_SPT)) |
| 532 | return 0; |
| 533 | |
| 534 | if (!r->cfg.gc_interval) |
| 535 | return 0; |
| 536 | |
| 537 | if (!r->cfg.expire) |
| 538 | return 0; |
| 539 | |
| 540 | /* This is the best we've got: We cannot release and re-grab lock, |
| 541 | * since checkentry() is called before ip_tables.c grabs ipt_mutex. |
| 542 | * We also cannot grab the hashtable spinlock, since htable_create will |
| 543 | * call vmalloc, and that can sleep. And we cannot just re-search |
| 544 | * the list of htable's in htable_create(), since then we would |
| 545 | * create duplicate proc files. -HW */ |
| 546 | down(&hlimit_mutex); |
| 547 | r->hinfo = htable_find_get(r->name); |
| 548 | if (!r->hinfo && (htable_create(r) != 0)) { |
| 549 | up(&hlimit_mutex); |
| 550 | return 0; |
| 551 | } |
| 552 | up(&hlimit_mutex); |
| 553 | |
| 554 | /* Ugly hack: For SMP, we only want to use one set */ |
| 555 | r->u.master = r; |
| 556 | |
| 557 | return 1; |
| 558 | } |
| 559 | |
| 560 | static void |
| 561 | hashlimit_destroy(void *matchinfo, unsigned int matchsize) |
| 562 | { |
| 563 | struct ipt_hashlimit_info *r = (struct ipt_hashlimit_info *) matchinfo; |
| 564 | |
| 565 | htable_put(r->hinfo); |
| 566 | } |
| 567 | |
| 568 | static struct ipt_match ipt_hashlimit = { |
| 569 | .name = "hashlimit", |
| 570 | .match = hashlimit_match, |
| 571 | .checkentry = hashlimit_checkentry, |
| 572 | .destroy = hashlimit_destroy, |
| 573 | .me = THIS_MODULE |
| 574 | }; |
| 575 | |
| 576 | /* PROC stuff */ |
| 577 | |
| 578 | static void *dl_seq_start(struct seq_file *s, loff_t *pos) |
| 579 | { |
| 580 | struct proc_dir_entry *pde = s->private; |
| 581 | struct ipt_hashlimit_htable *htable = pde->data; |
| 582 | unsigned int *bucket; |
| 583 | |
| 584 | spin_lock_bh(&htable->lock); |
| 585 | if (*pos >= htable->cfg.size) |
| 586 | return NULL; |
| 587 | |
| 588 | bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC); |
| 589 | if (!bucket) |
| 590 | return ERR_PTR(-ENOMEM); |
| 591 | |
| 592 | *bucket = *pos; |
| 593 | return bucket; |
| 594 | } |
| 595 | |
| 596 | static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos) |
| 597 | { |
| 598 | struct proc_dir_entry *pde = s->private; |
| 599 | struct ipt_hashlimit_htable *htable = pde->data; |
| 600 | unsigned int *bucket = (unsigned int *)v; |
| 601 | |
| 602 | *pos = ++(*bucket); |
| 603 | if (*pos >= htable->cfg.size) { |
| 604 | kfree(v); |
| 605 | return NULL; |
| 606 | } |
| 607 | return bucket; |
| 608 | } |
| 609 | |
| 610 | static void dl_seq_stop(struct seq_file *s, void *v) |
| 611 | { |
| 612 | struct proc_dir_entry *pde = s->private; |
| 613 | struct ipt_hashlimit_htable *htable = pde->data; |
| 614 | unsigned int *bucket = (unsigned int *)v; |
| 615 | |
| 616 | kfree(bucket); |
| 617 | |
| 618 | spin_unlock_bh(&htable->lock); |
| 619 | } |
| 620 | |
| 621 | static inline int dl_seq_real_show(struct dsthash_ent *ent, struct seq_file *s) |
| 622 | { |
| 623 | /* recalculate to show accurate numbers */ |
| 624 | rateinfo_recalc(ent, jiffies); |
| 625 | |
| 626 | return seq_printf(s, "%ld %u.%u.%u.%u:%u->%u.%u.%u.%u:%u %u %u %u\n", |
| 627 | (long)(ent->expires - jiffies)/HZ, |
| 628 | NIPQUAD(ent->dst.src_ip), ntohs(ent->dst.src_port), |
| 629 | NIPQUAD(ent->dst.dst_ip), ntohs(ent->dst.dst_port), |
| 630 | ent->rateinfo.credit, ent->rateinfo.credit_cap, |
| 631 | ent->rateinfo.cost); |
| 632 | } |
| 633 | |
| 634 | static int dl_seq_show(struct seq_file *s, void *v) |
| 635 | { |
| 636 | struct proc_dir_entry *pde = s->private; |
| 637 | struct ipt_hashlimit_htable *htable = pde->data; |
| 638 | unsigned int *bucket = (unsigned int *)v; |
| 639 | struct dsthash_ent *ent; |
| 640 | struct hlist_node *pos; |
| 641 | |
| 642 | if (!hlist_empty(&htable->hash[*bucket])) |
| 643 | hlist_for_each_entry(ent, pos, &htable->hash[*bucket], node) { |
| 644 | if (dl_seq_real_show(ent, s)) { |
| 645 | /* buffer was filled and unable to print that tuple */ |
| 646 | return 1; |
| 647 | } |
| 648 | } |
| 649 | |
| 650 | return 0; |
| 651 | } |
| 652 | |
| 653 | static struct seq_operations dl_seq_ops = { |
| 654 | .start = dl_seq_start, |
| 655 | .next = dl_seq_next, |
| 656 | .stop = dl_seq_stop, |
| 657 | .show = dl_seq_show |
| 658 | }; |
| 659 | |
| 660 | static int dl_proc_open(struct inode *inode, struct file *file) |
| 661 | { |
| 662 | int ret = seq_open(file, &dl_seq_ops); |
| 663 | |
| 664 | if (!ret) { |
| 665 | struct seq_file *sf = file->private_data; |
| 666 | sf->private = PDE(inode); |
| 667 | } |
| 668 | return ret; |
| 669 | } |
| 670 | |
| 671 | static struct file_operations dl_file_ops = { |
| 672 | .owner = THIS_MODULE, |
| 673 | .open = dl_proc_open, |
| 674 | .read = seq_read, |
| 675 | .llseek = seq_lseek, |
| 676 | .release = seq_release |
| 677 | }; |
| 678 | |
| 679 | static int init_or_fini(int fini) |
| 680 | { |
| 681 | int ret = 0; |
| 682 | |
| 683 | if (fini) |
| 684 | goto cleanup; |
| 685 | |
| 686 | if (ipt_register_match(&ipt_hashlimit)) { |
| 687 | ret = -EINVAL; |
| 688 | goto cleanup_nothing; |
| 689 | } |
| 690 | |
| 691 | hashlimit_cachep = kmem_cache_create("ipt_hashlimit", |
| 692 | sizeof(struct dsthash_ent), 0, |
| 693 | 0, NULL, NULL); |
| 694 | if (!hashlimit_cachep) { |
| 695 | printk(KERN_ERR "Unable to create ipt_hashlimit slab cache\n"); |
| 696 | ret = -ENOMEM; |
| 697 | goto cleanup_unreg_match; |
| 698 | } |
| 699 | |
| 700 | hashlimit_procdir = proc_mkdir("ipt_hashlimit", proc_net); |
| 701 | if (!hashlimit_procdir) { |
| 702 | printk(KERN_ERR "Unable to create proc dir entry\n"); |
| 703 | ret = -ENOMEM; |
| 704 | goto cleanup_free_slab; |
| 705 | } |
| 706 | |
| 707 | return ret; |
| 708 | |
| 709 | cleanup: |
| 710 | remove_proc_entry("ipt_hashlimit", proc_net); |
| 711 | cleanup_free_slab: |
| 712 | kmem_cache_destroy(hashlimit_cachep); |
| 713 | cleanup_unreg_match: |
| 714 | ipt_unregister_match(&ipt_hashlimit); |
| 715 | cleanup_nothing: |
| 716 | return ret; |
| 717 | |
| 718 | } |
| 719 | |
| 720 | static int __init init(void) |
| 721 | { |
| 722 | return init_or_fini(0); |
| 723 | } |
| 724 | |
| 725 | static void __exit fini(void) |
| 726 | { |
| 727 | init_or_fini(1); |
| 728 | } |
| 729 | |
| 730 | module_init(init); |
| 731 | module_exit(fini); |