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