| 1 | /* |
| 2 | * xt_hashlimit - Netfilter module to limit the number of packets per time |
| 3 | * separately for each hashbucket (sourceip/sourceport/dstip/dstport) |
| 4 | * |
| 5 | * (C) 2003-2004 by Harald Welte <laforge@netfilter.org> |
| 6 | * (C) 2006-2012 Patrick McHardy <kaber@trash.net> |
| 7 | * Copyright © CC Computer Consultants GmbH, 2007 - 2008 |
| 8 | * |
| 9 | * Development of this code was funded by Astaro AG, http://www.astaro.com/ |
| 10 | */ |
| 11 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
| 12 | #include <linux/module.h> |
| 13 | #include <linux/spinlock.h> |
| 14 | #include <linux/random.h> |
| 15 | #include <linux/jhash.h> |
| 16 | #include <linux/slab.h> |
| 17 | #include <linux/vmalloc.h> |
| 18 | #include <linux/proc_fs.h> |
| 19 | #include <linux/seq_file.h> |
| 20 | #include <linux/list.h> |
| 21 | #include <linux/skbuff.h> |
| 22 | #include <linux/mm.h> |
| 23 | #include <linux/in.h> |
| 24 | #include <linux/ip.h> |
| 25 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) |
| 26 | #include <linux/ipv6.h> |
| 27 | #include <net/ipv6.h> |
| 28 | #endif |
| 29 | |
| 30 | #include <net/net_namespace.h> |
| 31 | #include <net/netns/generic.h> |
| 32 | |
| 33 | #include <linux/netfilter/x_tables.h> |
| 34 | #include <linux/netfilter_ipv4/ip_tables.h> |
| 35 | #include <linux/netfilter_ipv6/ip6_tables.h> |
| 36 | #include <linux/netfilter/xt_hashlimit.h> |
| 37 | #include <linux/mutex.h> |
| 38 | |
| 39 | MODULE_LICENSE("GPL"); |
| 40 | MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>"); |
| 41 | MODULE_AUTHOR("Jan Engelhardt <jengelh@medozas.de>"); |
| 42 | MODULE_DESCRIPTION("Xtables: per hash-bucket rate-limit match"); |
| 43 | MODULE_ALIAS("ipt_hashlimit"); |
| 44 | MODULE_ALIAS("ip6t_hashlimit"); |
| 45 | |
| 46 | struct hashlimit_net { |
| 47 | struct hlist_head htables; |
| 48 | struct proc_dir_entry *ipt_hashlimit; |
| 49 | struct proc_dir_entry *ip6t_hashlimit; |
| 50 | }; |
| 51 | |
| 52 | static int hashlimit_net_id; |
| 53 | static inline struct hashlimit_net *hashlimit_pernet(struct net *net) |
| 54 | { |
| 55 | return net_generic(net, hashlimit_net_id); |
| 56 | } |
| 57 | |
| 58 | /* need to declare this at the top */ |
| 59 | static const struct file_operations dl_file_ops; |
| 60 | |
| 61 | /* hash table crap */ |
| 62 | struct dsthash_dst { |
| 63 | union { |
| 64 | struct { |
| 65 | __be32 src; |
| 66 | __be32 dst; |
| 67 | } ip; |
| 68 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) |
| 69 | struct { |
| 70 | __be32 src[4]; |
| 71 | __be32 dst[4]; |
| 72 | } ip6; |
| 73 | #endif |
| 74 | }; |
| 75 | __be16 src_port; |
| 76 | __be16 dst_port; |
| 77 | }; |
| 78 | |
| 79 | struct dsthash_ent { |
| 80 | /* static / read-only parts in the beginning */ |
| 81 | struct hlist_node node; |
| 82 | struct dsthash_dst dst; |
| 83 | |
| 84 | /* modified structure members in the end */ |
| 85 | spinlock_t lock; |
| 86 | unsigned long expires; /* precalculated expiry time */ |
| 87 | struct { |
| 88 | unsigned long prev; /* last modification */ |
| 89 | u_int32_t credit; |
| 90 | u_int32_t credit_cap, cost; |
| 91 | } rateinfo; |
| 92 | struct rcu_head rcu; |
| 93 | }; |
| 94 | |
| 95 | struct xt_hashlimit_htable { |
| 96 | struct hlist_node node; /* global list of all htables */ |
| 97 | int use; |
| 98 | u_int8_t family; |
| 99 | bool rnd_initialized; |
| 100 | |
| 101 | struct hashlimit_cfg1 cfg; /* config */ |
| 102 | |
| 103 | /* used internally */ |
| 104 | spinlock_t lock; /* lock for list_head */ |
| 105 | u_int32_t rnd; /* random seed for hash */ |
| 106 | unsigned int count; /* number entries in table */ |
| 107 | struct delayed_work gc_work; |
| 108 | |
| 109 | /* seq_file stuff */ |
| 110 | struct proc_dir_entry *pde; |
| 111 | const char *name; |
| 112 | struct net *net; |
| 113 | |
| 114 | struct hlist_head hash[0]; /* hashtable itself */ |
| 115 | }; |
| 116 | |
| 117 | static DEFINE_MUTEX(hashlimit_mutex); /* protects htables list */ |
| 118 | static struct kmem_cache *hashlimit_cachep __read_mostly; |
| 119 | |
| 120 | static inline bool dst_cmp(const struct dsthash_ent *ent, |
| 121 | const struct dsthash_dst *b) |
| 122 | { |
| 123 | return !memcmp(&ent->dst, b, sizeof(ent->dst)); |
| 124 | } |
| 125 | |
| 126 | static u_int32_t |
| 127 | hash_dst(const struct xt_hashlimit_htable *ht, const struct dsthash_dst *dst) |
| 128 | { |
| 129 | u_int32_t hash = jhash2((const u32 *)dst, |
| 130 | sizeof(*dst)/sizeof(u32), |
| 131 | ht->rnd); |
| 132 | /* |
| 133 | * Instead of returning hash % ht->cfg.size (implying a divide) |
| 134 | * we return the high 32 bits of the (hash * ht->cfg.size) that will |
| 135 | * give results between [0 and cfg.size-1] and same hash distribution, |
| 136 | * but using a multiply, less expensive than a divide |
| 137 | */ |
| 138 | return reciprocal_scale(hash, ht->cfg.size); |
| 139 | } |
| 140 | |
| 141 | static struct dsthash_ent * |
| 142 | dsthash_find(const struct xt_hashlimit_htable *ht, |
| 143 | const struct dsthash_dst *dst) |
| 144 | { |
| 145 | struct dsthash_ent *ent; |
| 146 | u_int32_t hash = hash_dst(ht, dst); |
| 147 | |
| 148 | if (!hlist_empty(&ht->hash[hash])) { |
| 149 | hlist_for_each_entry_rcu(ent, &ht->hash[hash], node) |
| 150 | if (dst_cmp(ent, dst)) { |
| 151 | spin_lock(&ent->lock); |
| 152 | return ent; |
| 153 | } |
| 154 | } |
| 155 | return NULL; |
| 156 | } |
| 157 | |
| 158 | /* allocate dsthash_ent, initialize dst, put in htable and lock it */ |
| 159 | static struct dsthash_ent * |
| 160 | dsthash_alloc_init(struct xt_hashlimit_htable *ht, |
| 161 | const struct dsthash_dst *dst, bool *race) |
| 162 | { |
| 163 | struct dsthash_ent *ent; |
| 164 | |
| 165 | spin_lock(&ht->lock); |
| 166 | |
| 167 | /* Two or more packets may race to create the same entry in the |
| 168 | * hashtable, double check if this packet lost race. |
| 169 | */ |
| 170 | ent = dsthash_find(ht, dst); |
| 171 | if (ent != NULL) { |
| 172 | spin_unlock(&ht->lock); |
| 173 | *race = true; |
| 174 | return ent; |
| 175 | } |
| 176 | |
| 177 | /* initialize hash with random val at the time we allocate |
| 178 | * the first hashtable entry */ |
| 179 | if (unlikely(!ht->rnd_initialized)) { |
| 180 | get_random_bytes(&ht->rnd, sizeof(ht->rnd)); |
| 181 | ht->rnd_initialized = true; |
| 182 | } |
| 183 | |
| 184 | if (ht->cfg.max && ht->count >= ht->cfg.max) { |
| 185 | /* FIXME: do something. question is what.. */ |
| 186 | net_err_ratelimited("max count of %u reached\n", ht->cfg.max); |
| 187 | ent = NULL; |
| 188 | } else |
| 189 | ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC); |
| 190 | if (ent) { |
| 191 | memcpy(&ent->dst, dst, sizeof(ent->dst)); |
| 192 | spin_lock_init(&ent->lock); |
| 193 | |
| 194 | spin_lock(&ent->lock); |
| 195 | hlist_add_head_rcu(&ent->node, &ht->hash[hash_dst(ht, dst)]); |
| 196 | ht->count++; |
| 197 | } |
| 198 | spin_unlock(&ht->lock); |
| 199 | return ent; |
| 200 | } |
| 201 | |
| 202 | static void dsthash_free_rcu(struct rcu_head *head) |
| 203 | { |
| 204 | struct dsthash_ent *ent = container_of(head, struct dsthash_ent, rcu); |
| 205 | |
| 206 | kmem_cache_free(hashlimit_cachep, ent); |
| 207 | } |
| 208 | |
| 209 | static inline void |
| 210 | dsthash_free(struct xt_hashlimit_htable *ht, struct dsthash_ent *ent) |
| 211 | { |
| 212 | hlist_del_rcu(&ent->node); |
| 213 | call_rcu_bh(&ent->rcu, dsthash_free_rcu); |
| 214 | ht->count--; |
| 215 | } |
| 216 | static void htable_gc(struct work_struct *work); |
| 217 | |
| 218 | static int htable_create(struct net *net, struct xt_hashlimit_mtinfo1 *minfo, |
| 219 | u_int8_t family) |
| 220 | { |
| 221 | struct hashlimit_net *hashlimit_net = hashlimit_pernet(net); |
| 222 | struct xt_hashlimit_htable *hinfo; |
| 223 | unsigned int size; |
| 224 | unsigned int i; |
| 225 | |
| 226 | if (minfo->cfg.size) { |
| 227 | size = minfo->cfg.size; |
| 228 | } else { |
| 229 | size = (totalram_pages << PAGE_SHIFT) / 16384 / |
| 230 | sizeof(struct list_head); |
| 231 | if (totalram_pages > 1024 * 1024 * 1024 / PAGE_SIZE) |
| 232 | size = 8192; |
| 233 | if (size < 16) |
| 234 | size = 16; |
| 235 | } |
| 236 | /* FIXME: don't use vmalloc() here or anywhere else -HW */ |
| 237 | hinfo = vmalloc(sizeof(struct xt_hashlimit_htable) + |
| 238 | sizeof(struct list_head) * size); |
| 239 | if (hinfo == NULL) |
| 240 | return -ENOMEM; |
| 241 | minfo->hinfo = hinfo; |
| 242 | |
| 243 | /* copy match config into hashtable config */ |
| 244 | memcpy(&hinfo->cfg, &minfo->cfg, sizeof(hinfo->cfg)); |
| 245 | hinfo->cfg.size = size; |
| 246 | if (hinfo->cfg.max == 0) |
| 247 | hinfo->cfg.max = 8 * hinfo->cfg.size; |
| 248 | else if (hinfo->cfg.max < hinfo->cfg.size) |
| 249 | hinfo->cfg.max = hinfo->cfg.size; |
| 250 | |
| 251 | for (i = 0; i < hinfo->cfg.size; i++) |
| 252 | INIT_HLIST_HEAD(&hinfo->hash[i]); |
| 253 | |
| 254 | hinfo->use = 1; |
| 255 | hinfo->count = 0; |
| 256 | hinfo->family = family; |
| 257 | hinfo->rnd_initialized = false; |
| 258 | hinfo->name = kstrdup(minfo->name, GFP_KERNEL); |
| 259 | if (!hinfo->name) { |
| 260 | vfree(hinfo); |
| 261 | return -ENOMEM; |
| 262 | } |
| 263 | spin_lock_init(&hinfo->lock); |
| 264 | |
| 265 | hinfo->pde = proc_create_data(minfo->name, 0, |
| 266 | (family == NFPROTO_IPV4) ? |
| 267 | hashlimit_net->ipt_hashlimit : hashlimit_net->ip6t_hashlimit, |
| 268 | &dl_file_ops, hinfo); |
| 269 | if (hinfo->pde == NULL) { |
| 270 | kfree(hinfo->name); |
| 271 | vfree(hinfo); |
| 272 | return -ENOMEM; |
| 273 | } |
| 274 | hinfo->net = net; |
| 275 | |
| 276 | INIT_DEFERRABLE_WORK(&hinfo->gc_work, htable_gc); |
| 277 | queue_delayed_work(system_power_efficient_wq, &hinfo->gc_work, |
| 278 | msecs_to_jiffies(hinfo->cfg.gc_interval)); |
| 279 | |
| 280 | hlist_add_head(&hinfo->node, &hashlimit_net->htables); |
| 281 | |
| 282 | return 0; |
| 283 | } |
| 284 | |
| 285 | static bool select_all(const struct xt_hashlimit_htable *ht, |
| 286 | const struct dsthash_ent *he) |
| 287 | { |
| 288 | return 1; |
| 289 | } |
| 290 | |
| 291 | static bool select_gc(const struct xt_hashlimit_htable *ht, |
| 292 | const struct dsthash_ent *he) |
| 293 | { |
| 294 | return time_after_eq(jiffies, he->expires); |
| 295 | } |
| 296 | |
| 297 | static void htable_selective_cleanup(struct xt_hashlimit_htable *ht, |
| 298 | bool (*select)(const struct xt_hashlimit_htable *ht, |
| 299 | const struct dsthash_ent *he)) |
| 300 | { |
| 301 | unsigned int i; |
| 302 | |
| 303 | for (i = 0; i < ht->cfg.size; i++) { |
| 304 | struct dsthash_ent *dh; |
| 305 | struct hlist_node *n; |
| 306 | |
| 307 | spin_lock_bh(&ht->lock); |
| 308 | hlist_for_each_entry_safe(dh, n, &ht->hash[i], node) { |
| 309 | if ((*select)(ht, dh)) |
| 310 | dsthash_free(ht, dh); |
| 311 | } |
| 312 | spin_unlock_bh(&ht->lock); |
| 313 | cond_resched(); |
| 314 | } |
| 315 | } |
| 316 | |
| 317 | static void htable_gc(struct work_struct *work) |
| 318 | { |
| 319 | struct xt_hashlimit_htable *ht; |
| 320 | |
| 321 | ht = container_of(work, struct xt_hashlimit_htable, gc_work.work); |
| 322 | |
| 323 | htable_selective_cleanup(ht, select_gc); |
| 324 | |
| 325 | queue_delayed_work(system_power_efficient_wq, |
| 326 | &ht->gc_work, msecs_to_jiffies(ht->cfg.gc_interval)); |
| 327 | } |
| 328 | |
| 329 | static void htable_remove_proc_entry(struct xt_hashlimit_htable *hinfo) |
| 330 | { |
| 331 | struct hashlimit_net *hashlimit_net = hashlimit_pernet(hinfo->net); |
| 332 | struct proc_dir_entry *parent; |
| 333 | |
| 334 | if (hinfo->family == NFPROTO_IPV4) |
| 335 | parent = hashlimit_net->ipt_hashlimit; |
| 336 | else |
| 337 | parent = hashlimit_net->ip6t_hashlimit; |
| 338 | |
| 339 | if (parent != NULL) |
| 340 | remove_proc_entry(hinfo->name, parent); |
| 341 | } |
| 342 | |
| 343 | static void htable_destroy(struct xt_hashlimit_htable *hinfo) |
| 344 | { |
| 345 | cancel_delayed_work_sync(&hinfo->gc_work); |
| 346 | htable_remove_proc_entry(hinfo); |
| 347 | htable_selective_cleanup(hinfo, select_all); |
| 348 | kfree(hinfo->name); |
| 349 | vfree(hinfo); |
| 350 | } |
| 351 | |
| 352 | static struct xt_hashlimit_htable *htable_find_get(struct net *net, |
| 353 | const char *name, |
| 354 | u_int8_t family) |
| 355 | { |
| 356 | struct hashlimit_net *hashlimit_net = hashlimit_pernet(net); |
| 357 | struct xt_hashlimit_htable *hinfo; |
| 358 | |
| 359 | hlist_for_each_entry(hinfo, &hashlimit_net->htables, node) { |
| 360 | if (!strcmp(name, hinfo->name) && |
| 361 | hinfo->family == family) { |
| 362 | hinfo->use++; |
| 363 | return hinfo; |
| 364 | } |
| 365 | } |
| 366 | return NULL; |
| 367 | } |
| 368 | |
| 369 | static void htable_put(struct xt_hashlimit_htable *hinfo) |
| 370 | { |
| 371 | mutex_lock(&hashlimit_mutex); |
| 372 | if (--hinfo->use == 0) { |
| 373 | hlist_del(&hinfo->node); |
| 374 | htable_destroy(hinfo); |
| 375 | } |
| 376 | mutex_unlock(&hashlimit_mutex); |
| 377 | } |
| 378 | |
| 379 | /* The algorithm used is the Simple Token Bucket Filter (TBF) |
| 380 | * see net/sched/sch_tbf.c in the linux source tree |
| 381 | */ |
| 382 | |
| 383 | /* Rusty: This is my (non-mathematically-inclined) understanding of |
| 384 | this algorithm. The `average rate' in jiffies becomes your initial |
| 385 | amount of credit `credit' and the most credit you can ever have |
| 386 | `credit_cap'. The `peak rate' becomes the cost of passing the |
| 387 | test, `cost'. |
| 388 | |
| 389 | `prev' tracks the last packet hit: you gain one credit per jiffy. |
| 390 | If you get credit balance more than this, the extra credit is |
| 391 | discarded. Every time the match passes, you lose `cost' credits; |
| 392 | if you don't have that many, the test fails. |
| 393 | |
| 394 | See Alexey's formal explanation in net/sched/sch_tbf.c. |
| 395 | |
| 396 | To get the maximum range, we multiply by this factor (ie. you get N |
| 397 | credits per jiffy). We want to allow a rate as low as 1 per day |
| 398 | (slowest userspace tool allows), which means |
| 399 | CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie. |
| 400 | */ |
| 401 | #define MAX_CPJ (0xFFFFFFFF / (HZ*60*60*24)) |
| 402 | |
| 403 | /* Repeated shift and or gives us all 1s, final shift and add 1 gives |
| 404 | * us the power of 2 below the theoretical max, so GCC simply does a |
| 405 | * shift. */ |
| 406 | #define _POW2_BELOW2(x) ((x)|((x)>>1)) |
| 407 | #define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2)) |
| 408 | #define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4)) |
| 409 | #define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8)) |
| 410 | #define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16)) |
| 411 | #define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1) |
| 412 | |
| 413 | #define CREDITS_PER_JIFFY POW2_BELOW32(MAX_CPJ) |
| 414 | |
| 415 | /* in byte mode, the lowest possible rate is one packet/second. |
| 416 | * credit_cap is used as a counter that tells us how many times we can |
| 417 | * refill the "credits available" counter when it becomes empty. |
| 418 | */ |
| 419 | #define MAX_CPJ_BYTES (0xFFFFFFFF / HZ) |
| 420 | #define CREDITS_PER_JIFFY_BYTES POW2_BELOW32(MAX_CPJ_BYTES) |
| 421 | |
| 422 | static u32 xt_hashlimit_len_to_chunks(u32 len) |
| 423 | { |
| 424 | return (len >> XT_HASHLIMIT_BYTE_SHIFT) + 1; |
| 425 | } |
| 426 | |
| 427 | /* Precision saver. */ |
| 428 | static u32 user2credits(u32 user) |
| 429 | { |
| 430 | /* If multiplying would overflow... */ |
| 431 | if (user > 0xFFFFFFFF / (HZ*CREDITS_PER_JIFFY)) |
| 432 | /* Divide first. */ |
| 433 | return (user / XT_HASHLIMIT_SCALE) * HZ * CREDITS_PER_JIFFY; |
| 434 | |
| 435 | return (user * HZ * CREDITS_PER_JIFFY) / XT_HASHLIMIT_SCALE; |
| 436 | } |
| 437 | |
| 438 | static u32 user2credits_byte(u32 user) |
| 439 | { |
| 440 | u64 us = user; |
| 441 | us *= HZ * CREDITS_PER_JIFFY_BYTES; |
| 442 | return (u32) (us >> 32); |
| 443 | } |
| 444 | |
| 445 | static void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now, u32 mode) |
| 446 | { |
| 447 | unsigned long delta = now - dh->rateinfo.prev; |
| 448 | u32 cap; |
| 449 | |
| 450 | if (delta == 0) |
| 451 | return; |
| 452 | |
| 453 | dh->rateinfo.prev = now; |
| 454 | |
| 455 | if (mode & XT_HASHLIMIT_BYTES) { |
| 456 | u32 tmp = dh->rateinfo.credit; |
| 457 | dh->rateinfo.credit += CREDITS_PER_JIFFY_BYTES * delta; |
| 458 | cap = CREDITS_PER_JIFFY_BYTES * HZ; |
| 459 | if (tmp >= dh->rateinfo.credit) {/* overflow */ |
| 460 | dh->rateinfo.credit = cap; |
| 461 | return; |
| 462 | } |
| 463 | } else { |
| 464 | dh->rateinfo.credit += delta * CREDITS_PER_JIFFY; |
| 465 | cap = dh->rateinfo.credit_cap; |
| 466 | } |
| 467 | if (dh->rateinfo.credit > cap) |
| 468 | dh->rateinfo.credit = cap; |
| 469 | } |
| 470 | |
| 471 | static void rateinfo_init(struct dsthash_ent *dh, |
| 472 | struct xt_hashlimit_htable *hinfo) |
| 473 | { |
| 474 | dh->rateinfo.prev = jiffies; |
| 475 | if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) { |
| 476 | dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ; |
| 477 | dh->rateinfo.cost = user2credits_byte(hinfo->cfg.avg); |
| 478 | dh->rateinfo.credit_cap = hinfo->cfg.burst; |
| 479 | } else { |
| 480 | dh->rateinfo.credit = user2credits(hinfo->cfg.avg * |
| 481 | hinfo->cfg.burst); |
| 482 | dh->rateinfo.cost = user2credits(hinfo->cfg.avg); |
| 483 | dh->rateinfo.credit_cap = dh->rateinfo.credit; |
| 484 | } |
| 485 | } |
| 486 | |
| 487 | static inline __be32 maskl(__be32 a, unsigned int l) |
| 488 | { |
| 489 | return l ? htonl(ntohl(a) & ~0 << (32 - l)) : 0; |
| 490 | } |
| 491 | |
| 492 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) |
| 493 | static void hashlimit_ipv6_mask(__be32 *i, unsigned int p) |
| 494 | { |
| 495 | switch (p) { |
| 496 | case 0 ... 31: |
| 497 | i[0] = maskl(i[0], p); |
| 498 | i[1] = i[2] = i[3] = 0; |
| 499 | break; |
| 500 | case 32 ... 63: |
| 501 | i[1] = maskl(i[1], p - 32); |
| 502 | i[2] = i[3] = 0; |
| 503 | break; |
| 504 | case 64 ... 95: |
| 505 | i[2] = maskl(i[2], p - 64); |
| 506 | i[3] = 0; |
| 507 | break; |
| 508 | case 96 ... 127: |
| 509 | i[3] = maskl(i[3], p - 96); |
| 510 | break; |
| 511 | case 128: |
| 512 | break; |
| 513 | } |
| 514 | } |
| 515 | #endif |
| 516 | |
| 517 | static int |
| 518 | hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo, |
| 519 | struct dsthash_dst *dst, |
| 520 | const struct sk_buff *skb, unsigned int protoff) |
| 521 | { |
| 522 | __be16 _ports[2], *ports; |
| 523 | u8 nexthdr; |
| 524 | int poff; |
| 525 | |
| 526 | memset(dst, 0, sizeof(*dst)); |
| 527 | |
| 528 | switch (hinfo->family) { |
| 529 | case NFPROTO_IPV4: |
| 530 | if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) |
| 531 | dst->ip.dst = maskl(ip_hdr(skb)->daddr, |
| 532 | hinfo->cfg.dstmask); |
| 533 | if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) |
| 534 | dst->ip.src = maskl(ip_hdr(skb)->saddr, |
| 535 | hinfo->cfg.srcmask); |
| 536 | |
| 537 | if (!(hinfo->cfg.mode & |
| 538 | (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT))) |
| 539 | return 0; |
| 540 | nexthdr = ip_hdr(skb)->protocol; |
| 541 | break; |
| 542 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) |
| 543 | case NFPROTO_IPV6: |
| 544 | { |
| 545 | __be16 frag_off; |
| 546 | |
| 547 | if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) { |
| 548 | memcpy(&dst->ip6.dst, &ipv6_hdr(skb)->daddr, |
| 549 | sizeof(dst->ip6.dst)); |
| 550 | hashlimit_ipv6_mask(dst->ip6.dst, hinfo->cfg.dstmask); |
| 551 | } |
| 552 | if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) { |
| 553 | memcpy(&dst->ip6.src, &ipv6_hdr(skb)->saddr, |
| 554 | sizeof(dst->ip6.src)); |
| 555 | hashlimit_ipv6_mask(dst->ip6.src, hinfo->cfg.srcmask); |
| 556 | } |
| 557 | |
| 558 | if (!(hinfo->cfg.mode & |
| 559 | (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT))) |
| 560 | return 0; |
| 561 | nexthdr = ipv6_hdr(skb)->nexthdr; |
| 562 | protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr, &frag_off); |
| 563 | if ((int)protoff < 0) |
| 564 | return -1; |
| 565 | break; |
| 566 | } |
| 567 | #endif |
| 568 | default: |
| 569 | BUG(); |
| 570 | return 0; |
| 571 | } |
| 572 | |
| 573 | poff = proto_ports_offset(nexthdr); |
| 574 | if (poff >= 0) { |
| 575 | ports = skb_header_pointer(skb, protoff + poff, sizeof(_ports), |
| 576 | &_ports); |
| 577 | } else { |
| 578 | _ports[0] = _ports[1] = 0; |
| 579 | ports = _ports; |
| 580 | } |
| 581 | if (!ports) |
| 582 | return -1; |
| 583 | if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SPT) |
| 584 | dst->src_port = ports[0]; |
| 585 | if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DPT) |
| 586 | dst->dst_port = ports[1]; |
| 587 | return 0; |
| 588 | } |
| 589 | |
| 590 | static u32 hashlimit_byte_cost(unsigned int len, struct dsthash_ent *dh) |
| 591 | { |
| 592 | u64 tmp = xt_hashlimit_len_to_chunks(len); |
| 593 | tmp = tmp * dh->rateinfo.cost; |
| 594 | |
| 595 | if (unlikely(tmp > CREDITS_PER_JIFFY_BYTES * HZ)) |
| 596 | tmp = CREDITS_PER_JIFFY_BYTES * HZ; |
| 597 | |
| 598 | if (dh->rateinfo.credit < tmp && dh->rateinfo.credit_cap) { |
| 599 | dh->rateinfo.credit_cap--; |
| 600 | dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ; |
| 601 | } |
| 602 | return (u32) tmp; |
| 603 | } |
| 604 | |
| 605 | static bool |
| 606 | hashlimit_mt(const struct sk_buff *skb, struct xt_action_param *par) |
| 607 | { |
| 608 | const struct xt_hashlimit_mtinfo1 *info = par->matchinfo; |
| 609 | struct xt_hashlimit_htable *hinfo = info->hinfo; |
| 610 | unsigned long now = jiffies; |
| 611 | struct dsthash_ent *dh; |
| 612 | struct dsthash_dst dst; |
| 613 | bool race = false; |
| 614 | u32 cost; |
| 615 | |
| 616 | if (hashlimit_init_dst(hinfo, &dst, skb, par->thoff) < 0) |
| 617 | goto hotdrop; |
| 618 | |
| 619 | rcu_read_lock_bh(); |
| 620 | dh = dsthash_find(hinfo, &dst); |
| 621 | if (dh == NULL) { |
| 622 | dh = dsthash_alloc_init(hinfo, &dst, &race); |
| 623 | if (dh == NULL) { |
| 624 | rcu_read_unlock_bh(); |
| 625 | goto hotdrop; |
| 626 | } else if (race) { |
| 627 | /* Already got an entry, update expiration timeout */ |
| 628 | dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire); |
| 629 | rateinfo_recalc(dh, now, hinfo->cfg.mode); |
| 630 | } else { |
| 631 | dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire); |
| 632 | rateinfo_init(dh, hinfo); |
| 633 | } |
| 634 | } else { |
| 635 | /* update expiration timeout */ |
| 636 | dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire); |
| 637 | rateinfo_recalc(dh, now, hinfo->cfg.mode); |
| 638 | } |
| 639 | |
| 640 | if (info->cfg.mode & XT_HASHLIMIT_BYTES) |
| 641 | cost = hashlimit_byte_cost(skb->len, dh); |
| 642 | else |
| 643 | cost = dh->rateinfo.cost; |
| 644 | |
| 645 | if (dh->rateinfo.credit >= cost) { |
| 646 | /* below the limit */ |
| 647 | dh->rateinfo.credit -= cost; |
| 648 | spin_unlock(&dh->lock); |
| 649 | rcu_read_unlock_bh(); |
| 650 | return !(info->cfg.mode & XT_HASHLIMIT_INVERT); |
| 651 | } |
| 652 | |
| 653 | spin_unlock(&dh->lock); |
| 654 | rcu_read_unlock_bh(); |
| 655 | /* default match is underlimit - so over the limit, we need to invert */ |
| 656 | return info->cfg.mode & XT_HASHLIMIT_INVERT; |
| 657 | |
| 658 | hotdrop: |
| 659 | par->hotdrop = true; |
| 660 | return false; |
| 661 | } |
| 662 | |
| 663 | static int hashlimit_mt_check(const struct xt_mtchk_param *par) |
| 664 | { |
| 665 | struct net *net = par->net; |
| 666 | struct xt_hashlimit_mtinfo1 *info = par->matchinfo; |
| 667 | int ret; |
| 668 | |
| 669 | if (info->cfg.gc_interval == 0 || info->cfg.expire == 0) |
| 670 | return -EINVAL; |
| 671 | if (info->name[sizeof(info->name)-1] != '\0') |
| 672 | return -EINVAL; |
| 673 | if (par->family == NFPROTO_IPV4) { |
| 674 | if (info->cfg.srcmask > 32 || info->cfg.dstmask > 32) |
| 675 | return -EINVAL; |
| 676 | } else { |
| 677 | if (info->cfg.srcmask > 128 || info->cfg.dstmask > 128) |
| 678 | return -EINVAL; |
| 679 | } |
| 680 | |
| 681 | if (info->cfg.mode & ~XT_HASHLIMIT_ALL) { |
| 682 | pr_info("Unknown mode mask %X, kernel too old?\n", |
| 683 | info->cfg.mode); |
| 684 | return -EINVAL; |
| 685 | } |
| 686 | |
| 687 | /* Check for overflow. */ |
| 688 | if (info->cfg.mode & XT_HASHLIMIT_BYTES) { |
| 689 | if (user2credits_byte(info->cfg.avg) == 0) { |
| 690 | pr_info("overflow, rate too high: %u\n", info->cfg.avg); |
| 691 | return -EINVAL; |
| 692 | } |
| 693 | } else if (info->cfg.burst == 0 || |
| 694 | user2credits(info->cfg.avg * info->cfg.burst) < |
| 695 | user2credits(info->cfg.avg)) { |
| 696 | pr_info("overflow, try lower: %u/%u\n", |
| 697 | info->cfg.avg, info->cfg.burst); |
| 698 | return -ERANGE; |
| 699 | } |
| 700 | |
| 701 | mutex_lock(&hashlimit_mutex); |
| 702 | info->hinfo = htable_find_get(net, info->name, par->family); |
| 703 | if (info->hinfo == NULL) { |
| 704 | ret = htable_create(net, info, par->family); |
| 705 | if (ret < 0) { |
| 706 | mutex_unlock(&hashlimit_mutex); |
| 707 | return ret; |
| 708 | } |
| 709 | } |
| 710 | mutex_unlock(&hashlimit_mutex); |
| 711 | return 0; |
| 712 | } |
| 713 | |
| 714 | static void hashlimit_mt_destroy(const struct xt_mtdtor_param *par) |
| 715 | { |
| 716 | const struct xt_hashlimit_mtinfo1 *info = par->matchinfo; |
| 717 | |
| 718 | htable_put(info->hinfo); |
| 719 | } |
| 720 | |
| 721 | static struct xt_match hashlimit_mt_reg[] __read_mostly = { |
| 722 | { |
| 723 | .name = "hashlimit", |
| 724 | .revision = 1, |
| 725 | .family = NFPROTO_IPV4, |
| 726 | .match = hashlimit_mt, |
| 727 | .matchsize = sizeof(struct xt_hashlimit_mtinfo1), |
| 728 | .checkentry = hashlimit_mt_check, |
| 729 | .destroy = hashlimit_mt_destroy, |
| 730 | .me = THIS_MODULE, |
| 731 | }, |
| 732 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) |
| 733 | { |
| 734 | .name = "hashlimit", |
| 735 | .revision = 1, |
| 736 | .family = NFPROTO_IPV6, |
| 737 | .match = hashlimit_mt, |
| 738 | .matchsize = sizeof(struct xt_hashlimit_mtinfo1), |
| 739 | .checkentry = hashlimit_mt_check, |
| 740 | .destroy = hashlimit_mt_destroy, |
| 741 | .me = THIS_MODULE, |
| 742 | }, |
| 743 | #endif |
| 744 | }; |
| 745 | |
| 746 | /* PROC stuff */ |
| 747 | static void *dl_seq_start(struct seq_file *s, loff_t *pos) |
| 748 | __acquires(htable->lock) |
| 749 | { |
| 750 | struct xt_hashlimit_htable *htable = s->private; |
| 751 | unsigned int *bucket; |
| 752 | |
| 753 | spin_lock_bh(&htable->lock); |
| 754 | if (*pos >= htable->cfg.size) |
| 755 | return NULL; |
| 756 | |
| 757 | bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC); |
| 758 | if (!bucket) |
| 759 | return ERR_PTR(-ENOMEM); |
| 760 | |
| 761 | *bucket = *pos; |
| 762 | return bucket; |
| 763 | } |
| 764 | |
| 765 | static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos) |
| 766 | { |
| 767 | struct xt_hashlimit_htable *htable = s->private; |
| 768 | unsigned int *bucket = (unsigned int *)v; |
| 769 | |
| 770 | *pos = ++(*bucket); |
| 771 | if (*pos >= htable->cfg.size) { |
| 772 | kfree(v); |
| 773 | return NULL; |
| 774 | } |
| 775 | return bucket; |
| 776 | } |
| 777 | |
| 778 | static void dl_seq_stop(struct seq_file *s, void *v) |
| 779 | __releases(htable->lock) |
| 780 | { |
| 781 | struct xt_hashlimit_htable *htable = s->private; |
| 782 | unsigned int *bucket = (unsigned int *)v; |
| 783 | |
| 784 | if (!IS_ERR(bucket)) |
| 785 | kfree(bucket); |
| 786 | spin_unlock_bh(&htable->lock); |
| 787 | } |
| 788 | |
| 789 | static int dl_seq_real_show(struct dsthash_ent *ent, u_int8_t family, |
| 790 | struct seq_file *s) |
| 791 | { |
| 792 | const struct xt_hashlimit_htable *ht = s->private; |
| 793 | |
| 794 | spin_lock(&ent->lock); |
| 795 | /* recalculate to show accurate numbers */ |
| 796 | rateinfo_recalc(ent, jiffies, ht->cfg.mode); |
| 797 | |
| 798 | switch (family) { |
| 799 | case NFPROTO_IPV4: |
| 800 | seq_printf(s, "%ld %pI4:%u->%pI4:%u %u %u %u\n", |
| 801 | (long)(ent->expires - jiffies)/HZ, |
| 802 | &ent->dst.ip.src, |
| 803 | ntohs(ent->dst.src_port), |
| 804 | &ent->dst.ip.dst, |
| 805 | ntohs(ent->dst.dst_port), |
| 806 | ent->rateinfo.credit, ent->rateinfo.credit_cap, |
| 807 | ent->rateinfo.cost); |
| 808 | break; |
| 809 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) |
| 810 | case NFPROTO_IPV6: |
| 811 | seq_printf(s, "%ld %pI6:%u->%pI6:%u %u %u %u\n", |
| 812 | (long)(ent->expires - jiffies)/HZ, |
| 813 | &ent->dst.ip6.src, |
| 814 | ntohs(ent->dst.src_port), |
| 815 | &ent->dst.ip6.dst, |
| 816 | ntohs(ent->dst.dst_port), |
| 817 | ent->rateinfo.credit, ent->rateinfo.credit_cap, |
| 818 | ent->rateinfo.cost); |
| 819 | break; |
| 820 | #endif |
| 821 | default: |
| 822 | BUG(); |
| 823 | } |
| 824 | spin_unlock(&ent->lock); |
| 825 | return seq_has_overflowed(s); |
| 826 | } |
| 827 | |
| 828 | static int dl_seq_show(struct seq_file *s, void *v) |
| 829 | { |
| 830 | struct xt_hashlimit_htable *htable = s->private; |
| 831 | unsigned int *bucket = (unsigned int *)v; |
| 832 | struct dsthash_ent *ent; |
| 833 | |
| 834 | if (!hlist_empty(&htable->hash[*bucket])) { |
| 835 | hlist_for_each_entry(ent, &htable->hash[*bucket], node) |
| 836 | if (dl_seq_real_show(ent, htable->family, s)) |
| 837 | return -1; |
| 838 | } |
| 839 | return 0; |
| 840 | } |
| 841 | |
| 842 | static const struct seq_operations dl_seq_ops = { |
| 843 | .start = dl_seq_start, |
| 844 | .next = dl_seq_next, |
| 845 | .stop = dl_seq_stop, |
| 846 | .show = dl_seq_show |
| 847 | }; |
| 848 | |
| 849 | static int dl_proc_open(struct inode *inode, struct file *file) |
| 850 | { |
| 851 | int ret = seq_open(file, &dl_seq_ops); |
| 852 | |
| 853 | if (!ret) { |
| 854 | struct seq_file *sf = file->private_data; |
| 855 | sf->private = PDE_DATA(inode); |
| 856 | } |
| 857 | return ret; |
| 858 | } |
| 859 | |
| 860 | static const struct file_operations dl_file_ops = { |
| 861 | .owner = THIS_MODULE, |
| 862 | .open = dl_proc_open, |
| 863 | .read = seq_read, |
| 864 | .llseek = seq_lseek, |
| 865 | .release = seq_release |
| 866 | }; |
| 867 | |
| 868 | static int __net_init hashlimit_proc_net_init(struct net *net) |
| 869 | { |
| 870 | struct hashlimit_net *hashlimit_net = hashlimit_pernet(net); |
| 871 | |
| 872 | hashlimit_net->ipt_hashlimit = proc_mkdir("ipt_hashlimit", net->proc_net); |
| 873 | if (!hashlimit_net->ipt_hashlimit) |
| 874 | return -ENOMEM; |
| 875 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) |
| 876 | hashlimit_net->ip6t_hashlimit = proc_mkdir("ip6t_hashlimit", net->proc_net); |
| 877 | if (!hashlimit_net->ip6t_hashlimit) { |
| 878 | remove_proc_entry("ipt_hashlimit", net->proc_net); |
| 879 | return -ENOMEM; |
| 880 | } |
| 881 | #endif |
| 882 | return 0; |
| 883 | } |
| 884 | |
| 885 | static void __net_exit hashlimit_proc_net_exit(struct net *net) |
| 886 | { |
| 887 | struct xt_hashlimit_htable *hinfo; |
| 888 | struct hashlimit_net *hashlimit_net = hashlimit_pernet(net); |
| 889 | |
| 890 | /* hashlimit_net_exit() is called before hashlimit_mt_destroy(). |
| 891 | * Make sure that the parent ipt_hashlimit and ip6t_hashlimit proc |
| 892 | * entries is empty before trying to remove it. |
| 893 | */ |
| 894 | mutex_lock(&hashlimit_mutex); |
| 895 | hlist_for_each_entry(hinfo, &hashlimit_net->htables, node) |
| 896 | htable_remove_proc_entry(hinfo); |
| 897 | hashlimit_net->ipt_hashlimit = NULL; |
| 898 | hashlimit_net->ip6t_hashlimit = NULL; |
| 899 | mutex_unlock(&hashlimit_mutex); |
| 900 | |
| 901 | remove_proc_entry("ipt_hashlimit", net->proc_net); |
| 902 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) |
| 903 | remove_proc_entry("ip6t_hashlimit", net->proc_net); |
| 904 | #endif |
| 905 | } |
| 906 | |
| 907 | static int __net_init hashlimit_net_init(struct net *net) |
| 908 | { |
| 909 | struct hashlimit_net *hashlimit_net = hashlimit_pernet(net); |
| 910 | |
| 911 | INIT_HLIST_HEAD(&hashlimit_net->htables); |
| 912 | return hashlimit_proc_net_init(net); |
| 913 | } |
| 914 | |
| 915 | static void __net_exit hashlimit_net_exit(struct net *net) |
| 916 | { |
| 917 | hashlimit_proc_net_exit(net); |
| 918 | } |
| 919 | |
| 920 | static struct pernet_operations hashlimit_net_ops = { |
| 921 | .init = hashlimit_net_init, |
| 922 | .exit = hashlimit_net_exit, |
| 923 | .id = &hashlimit_net_id, |
| 924 | .size = sizeof(struct hashlimit_net), |
| 925 | }; |
| 926 | |
| 927 | static int __init hashlimit_mt_init(void) |
| 928 | { |
| 929 | int err; |
| 930 | |
| 931 | err = register_pernet_subsys(&hashlimit_net_ops); |
| 932 | if (err < 0) |
| 933 | return err; |
| 934 | err = xt_register_matches(hashlimit_mt_reg, |
| 935 | ARRAY_SIZE(hashlimit_mt_reg)); |
| 936 | if (err < 0) |
| 937 | goto err1; |
| 938 | |
| 939 | err = -ENOMEM; |
| 940 | hashlimit_cachep = kmem_cache_create("xt_hashlimit", |
| 941 | sizeof(struct dsthash_ent), 0, 0, |
| 942 | NULL); |
| 943 | if (!hashlimit_cachep) { |
| 944 | pr_warn("unable to create slab cache\n"); |
| 945 | goto err2; |
| 946 | } |
| 947 | return 0; |
| 948 | |
| 949 | err2: |
| 950 | xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg)); |
| 951 | err1: |
| 952 | unregister_pernet_subsys(&hashlimit_net_ops); |
| 953 | return err; |
| 954 | |
| 955 | } |
| 956 | |
| 957 | static void __exit hashlimit_mt_exit(void) |
| 958 | { |
| 959 | xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg)); |
| 960 | unregister_pernet_subsys(&hashlimit_net_ops); |
| 961 | |
| 962 | rcu_barrier_bh(); |
| 963 | kmem_cache_destroy(hashlimit_cachep); |
| 964 | } |
| 965 | |
| 966 | module_init(hashlimit_mt_init); |
| 967 | module_exit(hashlimit_mt_exit); |