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 unsigned 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_v1; 60 static const struct file_operations dl_file_ops; 61 62 /* hash table crap */ 63 struct dsthash_dst { 64 union { 65 struct { 66 __be32 src; 67 __be32 dst; 68 } ip; 69 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) 70 struct { 71 __be32 src[4]; 72 __be32 dst[4]; 73 } ip6; 74 #endif 75 }; 76 __be16 src_port; 77 __be16 dst_port; 78 }; 79 80 struct dsthash_ent { 81 /* static / read-only parts in the beginning */ 82 struct hlist_node node; 83 struct dsthash_dst dst; 84 85 /* modified structure members in the end */ 86 spinlock_t lock; 87 unsigned long expires; /* precalculated expiry time */ 88 struct { 89 unsigned long prev; /* last modification */ 90 u_int64_t credit; 91 u_int64_t credit_cap, cost; 92 } rateinfo; 93 struct rcu_head rcu; 94 }; 95 96 struct xt_hashlimit_htable { 97 struct hlist_node node; /* global list of all htables */ 98 int use; 99 u_int8_t family; 100 bool rnd_initialized; 101 102 struct hashlimit_cfg2 cfg; /* config */ 103 104 /* used internally */ 105 spinlock_t lock; /* lock for list_head */ 106 u_int32_t rnd; /* random seed for hash */ 107 unsigned int count; /* number entries in table */ 108 struct delayed_work gc_work; 109 110 /* seq_file stuff */ 111 struct proc_dir_entry *pde; 112 const char *name; 113 struct net *net; 114 115 struct hlist_head hash[0]; /* hashtable itself */ 116 }; 117 118 static int 119 cfg_copy(struct hashlimit_cfg2 *to, void *from, int revision) 120 { 121 if (revision == 1) { 122 struct hashlimit_cfg1 *cfg = from; 123 124 to->mode = cfg->mode; 125 to->avg = cfg->avg; 126 to->burst = cfg->burst; 127 to->size = cfg->size; 128 to->max = cfg->max; 129 to->gc_interval = cfg->gc_interval; 130 to->expire = cfg->expire; 131 to->srcmask = cfg->srcmask; 132 to->dstmask = cfg->dstmask; 133 } else if (revision == 2) { 134 memcpy(to, from, sizeof(struct hashlimit_cfg2)); 135 } else { 136 return -EINVAL; 137 } 138 139 return 0; 140 } 141 142 static DEFINE_MUTEX(hashlimit_mutex); /* protects htables list */ 143 static struct kmem_cache *hashlimit_cachep __read_mostly; 144 145 static inline bool dst_cmp(const struct dsthash_ent *ent, 146 const struct dsthash_dst *b) 147 { 148 return !memcmp(&ent->dst, b, sizeof(ent->dst)); 149 } 150 151 static u_int32_t 152 hash_dst(const struct xt_hashlimit_htable *ht, const struct dsthash_dst *dst) 153 { 154 u_int32_t hash = jhash2((const u32 *)dst, 155 sizeof(*dst)/sizeof(u32), 156 ht->rnd); 157 /* 158 * Instead of returning hash % ht->cfg.size (implying a divide) 159 * we return the high 32 bits of the (hash * ht->cfg.size) that will 160 * give results between [0 and cfg.size-1] and same hash distribution, 161 * but using a multiply, less expensive than a divide 162 */ 163 return reciprocal_scale(hash, ht->cfg.size); 164 } 165 166 static struct dsthash_ent * 167 dsthash_find(const struct xt_hashlimit_htable *ht, 168 const struct dsthash_dst *dst) 169 { 170 struct dsthash_ent *ent; 171 u_int32_t hash = hash_dst(ht, dst); 172 173 if (!hlist_empty(&ht->hash[hash])) { 174 hlist_for_each_entry_rcu(ent, &ht->hash[hash], node) 175 if (dst_cmp(ent, dst)) { 176 spin_lock(&ent->lock); 177 return ent; 178 } 179 } 180 return NULL; 181 } 182 183 /* allocate dsthash_ent, initialize dst, put in htable and lock it */ 184 static struct dsthash_ent * 185 dsthash_alloc_init(struct xt_hashlimit_htable *ht, 186 const struct dsthash_dst *dst, bool *race) 187 { 188 struct dsthash_ent *ent; 189 190 spin_lock(&ht->lock); 191 192 /* Two or more packets may race to create the same entry in the 193 * hashtable, double check if this packet lost race. 194 */ 195 ent = dsthash_find(ht, dst); 196 if (ent != NULL) { 197 spin_unlock(&ht->lock); 198 *race = true; 199 return ent; 200 } 201 202 /* initialize hash with random val at the time we allocate 203 * the first hashtable entry */ 204 if (unlikely(!ht->rnd_initialized)) { 205 get_random_bytes(&ht->rnd, sizeof(ht->rnd)); 206 ht->rnd_initialized = true; 207 } 208 209 if (ht->cfg.max && ht->count >= ht->cfg.max) { 210 /* FIXME: do something. question is what.. */ 211 net_err_ratelimited("max count of %u reached\n", ht->cfg.max); 212 ent = NULL; 213 } else 214 ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC); 215 if (ent) { 216 memcpy(&ent->dst, dst, sizeof(ent->dst)); 217 spin_lock_init(&ent->lock); 218 219 spin_lock(&ent->lock); 220 hlist_add_head_rcu(&ent->node, &ht->hash[hash_dst(ht, dst)]); 221 ht->count++; 222 } 223 spin_unlock(&ht->lock); 224 return ent; 225 } 226 227 static void dsthash_free_rcu(struct rcu_head *head) 228 { 229 struct dsthash_ent *ent = container_of(head, struct dsthash_ent, rcu); 230 231 kmem_cache_free(hashlimit_cachep, ent); 232 } 233 234 static inline void 235 dsthash_free(struct xt_hashlimit_htable *ht, struct dsthash_ent *ent) 236 { 237 hlist_del_rcu(&ent->node); 238 call_rcu_bh(&ent->rcu, dsthash_free_rcu); 239 ht->count--; 240 } 241 static void htable_gc(struct work_struct *work); 242 243 static int htable_create(struct net *net, struct hashlimit_cfg2 *cfg, 244 const char *name, u_int8_t family, 245 struct xt_hashlimit_htable **out_hinfo, 246 int revision) 247 { 248 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net); 249 struct xt_hashlimit_htable *hinfo; 250 unsigned int size, i; 251 int ret; 252 253 if (cfg->size) { 254 size = cfg->size; 255 } else { 256 size = (totalram_pages << PAGE_SHIFT) / 16384 / 257 sizeof(struct list_head); 258 if (totalram_pages > 1024 * 1024 * 1024 / PAGE_SIZE) 259 size = 8192; 260 if (size < 16) 261 size = 16; 262 } 263 /* FIXME: don't use vmalloc() here or anywhere else -HW */ 264 hinfo = vmalloc(sizeof(struct xt_hashlimit_htable) + 265 sizeof(struct list_head) * size); 266 if (hinfo == NULL) 267 return -ENOMEM; 268 *out_hinfo = hinfo; 269 270 /* copy match config into hashtable config */ 271 ret = cfg_copy(&hinfo->cfg, (void *)cfg, 2); 272 273 if (ret) 274 return ret; 275 276 hinfo->cfg.size = size; 277 if (hinfo->cfg.max == 0) 278 hinfo->cfg.max = 8 * hinfo->cfg.size; 279 else if (hinfo->cfg.max < hinfo->cfg.size) 280 hinfo->cfg.max = hinfo->cfg.size; 281 282 for (i = 0; i < hinfo->cfg.size; i++) 283 INIT_HLIST_HEAD(&hinfo->hash[i]); 284 285 hinfo->use = 1; 286 hinfo->count = 0; 287 hinfo->family = family; 288 hinfo->rnd_initialized = false; 289 hinfo->name = kstrdup(name, GFP_KERNEL); 290 if (!hinfo->name) { 291 vfree(hinfo); 292 return -ENOMEM; 293 } 294 spin_lock_init(&hinfo->lock); 295 296 hinfo->pde = proc_create_data(name, 0, 297 (family == NFPROTO_IPV4) ? 298 hashlimit_net->ipt_hashlimit : hashlimit_net->ip6t_hashlimit, 299 (revision == 1) ? &dl_file_ops_v1 : &dl_file_ops, 300 hinfo); 301 if (hinfo->pde == NULL) { 302 kfree(hinfo->name); 303 vfree(hinfo); 304 return -ENOMEM; 305 } 306 hinfo->net = net; 307 308 INIT_DEFERRABLE_WORK(&hinfo->gc_work, htable_gc); 309 queue_delayed_work(system_power_efficient_wq, &hinfo->gc_work, 310 msecs_to_jiffies(hinfo->cfg.gc_interval)); 311 312 hlist_add_head(&hinfo->node, &hashlimit_net->htables); 313 314 return 0; 315 } 316 317 static bool select_all(const struct xt_hashlimit_htable *ht, 318 const struct dsthash_ent *he) 319 { 320 return 1; 321 } 322 323 static bool select_gc(const struct xt_hashlimit_htable *ht, 324 const struct dsthash_ent *he) 325 { 326 return time_after_eq(jiffies, he->expires); 327 } 328 329 static void htable_selective_cleanup(struct xt_hashlimit_htable *ht, 330 bool (*select)(const struct xt_hashlimit_htable *ht, 331 const struct dsthash_ent *he)) 332 { 333 unsigned int i; 334 335 for (i = 0; i < ht->cfg.size; i++) { 336 struct dsthash_ent *dh; 337 struct hlist_node *n; 338 339 spin_lock_bh(&ht->lock); 340 hlist_for_each_entry_safe(dh, n, &ht->hash[i], node) { 341 if ((*select)(ht, dh)) 342 dsthash_free(ht, dh); 343 } 344 spin_unlock_bh(&ht->lock); 345 cond_resched(); 346 } 347 } 348 349 static void htable_gc(struct work_struct *work) 350 { 351 struct xt_hashlimit_htable *ht; 352 353 ht = container_of(work, struct xt_hashlimit_htable, gc_work.work); 354 355 htable_selective_cleanup(ht, select_gc); 356 357 queue_delayed_work(system_power_efficient_wq, 358 &ht->gc_work, msecs_to_jiffies(ht->cfg.gc_interval)); 359 } 360 361 static void htable_remove_proc_entry(struct xt_hashlimit_htable *hinfo) 362 { 363 struct hashlimit_net *hashlimit_net = hashlimit_pernet(hinfo->net); 364 struct proc_dir_entry *parent; 365 366 if (hinfo->family == NFPROTO_IPV4) 367 parent = hashlimit_net->ipt_hashlimit; 368 else 369 parent = hashlimit_net->ip6t_hashlimit; 370 371 if (parent != NULL) 372 remove_proc_entry(hinfo->name, parent); 373 } 374 375 static void htable_destroy(struct xt_hashlimit_htable *hinfo) 376 { 377 cancel_delayed_work_sync(&hinfo->gc_work); 378 htable_remove_proc_entry(hinfo); 379 htable_selective_cleanup(hinfo, select_all); 380 kfree(hinfo->name); 381 vfree(hinfo); 382 } 383 384 static struct xt_hashlimit_htable *htable_find_get(struct net *net, 385 const char *name, 386 u_int8_t family) 387 { 388 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net); 389 struct xt_hashlimit_htable *hinfo; 390 391 hlist_for_each_entry(hinfo, &hashlimit_net->htables, node) { 392 if (!strcmp(name, hinfo->name) && 393 hinfo->family == family) { 394 hinfo->use++; 395 return hinfo; 396 } 397 } 398 return NULL; 399 } 400 401 static void htable_put(struct xt_hashlimit_htable *hinfo) 402 { 403 mutex_lock(&hashlimit_mutex); 404 if (--hinfo->use == 0) { 405 hlist_del(&hinfo->node); 406 htable_destroy(hinfo); 407 } 408 mutex_unlock(&hashlimit_mutex); 409 } 410 411 /* The algorithm used is the Simple Token Bucket Filter (TBF) 412 * see net/sched/sch_tbf.c in the linux source tree 413 */ 414 415 /* Rusty: This is my (non-mathematically-inclined) understanding of 416 this algorithm. The `average rate' in jiffies becomes your initial 417 amount of credit `credit' and the most credit you can ever have 418 `credit_cap'. The `peak rate' becomes the cost of passing the 419 test, `cost'. 420 421 `prev' tracks the last packet hit: you gain one credit per jiffy. 422 If you get credit balance more than this, the extra credit is 423 discarded. Every time the match passes, you lose `cost' credits; 424 if you don't have that many, the test fails. 425 426 See Alexey's formal explanation in net/sched/sch_tbf.c. 427 428 To get the maximum range, we multiply by this factor (ie. you get N 429 credits per jiffy). We want to allow a rate as low as 1 per day 430 (slowest userspace tool allows), which means 431 CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie. 432 */ 433 #define MAX_CPJ_v1 (0xFFFFFFFF / (HZ*60*60*24)) 434 #define MAX_CPJ (0xFFFFFFFFFFFFFFFFULL / (HZ*60*60*24)) 435 436 /* Repeated shift and or gives us all 1s, final shift and add 1 gives 437 * us the power of 2 below the theoretical max, so GCC simply does a 438 * shift. */ 439 #define _POW2_BELOW2(x) ((x)|((x)>>1)) 440 #define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2)) 441 #define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4)) 442 #define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8)) 443 #define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16)) 444 #define _POW2_BELOW64(x) (_POW2_BELOW32(x)|_POW2_BELOW32((x)>>32)) 445 #define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1) 446 #define POW2_BELOW64(x) ((_POW2_BELOW64(x)>>1) + 1) 447 448 #define CREDITS_PER_JIFFY POW2_BELOW64(MAX_CPJ) 449 #define CREDITS_PER_JIFFY_v1 POW2_BELOW32(MAX_CPJ_v1) 450 451 /* in byte mode, the lowest possible rate is one packet/second. 452 * credit_cap is used as a counter that tells us how many times we can 453 * refill the "credits available" counter when it becomes empty. 454 */ 455 #define MAX_CPJ_BYTES (0xFFFFFFFF / HZ) 456 #define CREDITS_PER_JIFFY_BYTES POW2_BELOW32(MAX_CPJ_BYTES) 457 458 static u32 xt_hashlimit_len_to_chunks(u32 len) 459 { 460 return (len >> XT_HASHLIMIT_BYTE_SHIFT) + 1; 461 } 462 463 /* Precision saver. */ 464 static u64 user2credits(u64 user, int revision) 465 { 466 u64 scale = (revision == 1) ? 467 XT_HASHLIMIT_SCALE : XT_HASHLIMIT_SCALE_v2; 468 u64 cpj = (revision == 1) ? 469 CREDITS_PER_JIFFY_v1 : CREDITS_PER_JIFFY; 470 471 /* Avoid overflow: divide the constant operands first */ 472 if (scale >= HZ * cpj) 473 return div64_u64(user, div64_u64(scale, HZ * cpj)); 474 475 return user * div64_u64(HZ * cpj, scale); 476 } 477 478 static u32 user2credits_byte(u32 user) 479 { 480 u64 us = user; 481 us *= HZ * CREDITS_PER_JIFFY_BYTES; 482 return (u32) (us >> 32); 483 } 484 485 static void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now, 486 u32 mode, int revision) 487 { 488 unsigned long delta = now - dh->rateinfo.prev; 489 u64 cap, cpj; 490 491 if (delta == 0) 492 return; 493 494 dh->rateinfo.prev = now; 495 496 if (mode & XT_HASHLIMIT_BYTES) { 497 u64 tmp = dh->rateinfo.credit; 498 dh->rateinfo.credit += CREDITS_PER_JIFFY_BYTES * delta; 499 cap = CREDITS_PER_JIFFY_BYTES * HZ; 500 if (tmp >= dh->rateinfo.credit) {/* overflow */ 501 dh->rateinfo.credit = cap; 502 return; 503 } 504 } else { 505 cpj = (revision == 1) ? 506 CREDITS_PER_JIFFY_v1 : CREDITS_PER_JIFFY; 507 dh->rateinfo.credit += delta * cpj; 508 cap = dh->rateinfo.credit_cap; 509 } 510 if (dh->rateinfo.credit > cap) 511 dh->rateinfo.credit = cap; 512 } 513 514 static void rateinfo_init(struct dsthash_ent *dh, 515 struct xt_hashlimit_htable *hinfo, int revision) 516 { 517 dh->rateinfo.prev = jiffies; 518 if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) { 519 dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ; 520 dh->rateinfo.cost = user2credits_byte(hinfo->cfg.avg); 521 dh->rateinfo.credit_cap = hinfo->cfg.burst; 522 } else { 523 dh->rateinfo.credit = user2credits(hinfo->cfg.avg * 524 hinfo->cfg.burst, revision); 525 dh->rateinfo.cost = user2credits(hinfo->cfg.avg, revision); 526 dh->rateinfo.credit_cap = dh->rateinfo.credit; 527 } 528 } 529 530 static inline __be32 maskl(__be32 a, unsigned int l) 531 { 532 return l ? htonl(ntohl(a) & ~0 << (32 - l)) : 0; 533 } 534 535 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) 536 static void hashlimit_ipv6_mask(__be32 *i, unsigned int p) 537 { 538 switch (p) { 539 case 0 ... 31: 540 i[0] = maskl(i[0], p); 541 i[1] = i[2] = i[3] = 0; 542 break; 543 case 32 ... 63: 544 i[1] = maskl(i[1], p - 32); 545 i[2] = i[3] = 0; 546 break; 547 case 64 ... 95: 548 i[2] = maskl(i[2], p - 64); 549 i[3] = 0; 550 break; 551 case 96 ... 127: 552 i[3] = maskl(i[3], p - 96); 553 break; 554 case 128: 555 break; 556 } 557 } 558 #endif 559 560 static int 561 hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo, 562 struct dsthash_dst *dst, 563 const struct sk_buff *skb, unsigned int protoff) 564 { 565 __be16 _ports[2], *ports; 566 u8 nexthdr; 567 int poff; 568 569 memset(dst, 0, sizeof(*dst)); 570 571 switch (hinfo->family) { 572 case NFPROTO_IPV4: 573 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) 574 dst->ip.dst = maskl(ip_hdr(skb)->daddr, 575 hinfo->cfg.dstmask); 576 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) 577 dst->ip.src = maskl(ip_hdr(skb)->saddr, 578 hinfo->cfg.srcmask); 579 580 if (!(hinfo->cfg.mode & 581 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT))) 582 return 0; 583 nexthdr = ip_hdr(skb)->protocol; 584 break; 585 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) 586 case NFPROTO_IPV6: 587 { 588 __be16 frag_off; 589 590 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) { 591 memcpy(&dst->ip6.dst, &ipv6_hdr(skb)->daddr, 592 sizeof(dst->ip6.dst)); 593 hashlimit_ipv6_mask(dst->ip6.dst, hinfo->cfg.dstmask); 594 } 595 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) { 596 memcpy(&dst->ip6.src, &ipv6_hdr(skb)->saddr, 597 sizeof(dst->ip6.src)); 598 hashlimit_ipv6_mask(dst->ip6.src, hinfo->cfg.srcmask); 599 } 600 601 if (!(hinfo->cfg.mode & 602 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT))) 603 return 0; 604 nexthdr = ipv6_hdr(skb)->nexthdr; 605 protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr, &frag_off); 606 if ((int)protoff < 0) 607 return -1; 608 break; 609 } 610 #endif 611 default: 612 BUG(); 613 return 0; 614 } 615 616 poff = proto_ports_offset(nexthdr); 617 if (poff >= 0) { 618 ports = skb_header_pointer(skb, protoff + poff, sizeof(_ports), 619 &_ports); 620 } else { 621 _ports[0] = _ports[1] = 0; 622 ports = _ports; 623 } 624 if (!ports) 625 return -1; 626 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SPT) 627 dst->src_port = ports[0]; 628 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DPT) 629 dst->dst_port = ports[1]; 630 return 0; 631 } 632 633 static u32 hashlimit_byte_cost(unsigned int len, struct dsthash_ent *dh) 634 { 635 u64 tmp = xt_hashlimit_len_to_chunks(len); 636 tmp = tmp * dh->rateinfo.cost; 637 638 if (unlikely(tmp > CREDITS_PER_JIFFY_BYTES * HZ)) 639 tmp = CREDITS_PER_JIFFY_BYTES * HZ; 640 641 if (dh->rateinfo.credit < tmp && dh->rateinfo.credit_cap) { 642 dh->rateinfo.credit_cap--; 643 dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ; 644 } 645 return (u32) tmp; 646 } 647 648 static bool 649 hashlimit_mt_common(const struct sk_buff *skb, struct xt_action_param *par, 650 struct xt_hashlimit_htable *hinfo, 651 const struct hashlimit_cfg2 *cfg, int revision) 652 { 653 unsigned long now = jiffies; 654 struct dsthash_ent *dh; 655 struct dsthash_dst dst; 656 bool race = false; 657 u64 cost; 658 659 if (hashlimit_init_dst(hinfo, &dst, skb, par->thoff) < 0) 660 goto hotdrop; 661 662 rcu_read_lock_bh(); 663 dh = dsthash_find(hinfo, &dst); 664 if (dh == NULL) { 665 dh = dsthash_alloc_init(hinfo, &dst, &race); 666 if (dh == NULL) { 667 rcu_read_unlock_bh(); 668 goto hotdrop; 669 } else if (race) { 670 /* Already got an entry, update expiration timeout */ 671 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire); 672 rateinfo_recalc(dh, now, hinfo->cfg.mode, revision); 673 } else { 674 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire); 675 rateinfo_init(dh, hinfo, revision); 676 } 677 } else { 678 /* update expiration timeout */ 679 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire); 680 rateinfo_recalc(dh, now, hinfo->cfg.mode, revision); 681 } 682 683 if (cfg->mode & XT_HASHLIMIT_BYTES) 684 cost = hashlimit_byte_cost(skb->len, dh); 685 else 686 cost = dh->rateinfo.cost; 687 688 if (dh->rateinfo.credit >= cost) { 689 /* below the limit */ 690 dh->rateinfo.credit -= cost; 691 spin_unlock(&dh->lock); 692 rcu_read_unlock_bh(); 693 return !(cfg->mode & XT_HASHLIMIT_INVERT); 694 } 695 696 spin_unlock(&dh->lock); 697 rcu_read_unlock_bh(); 698 /* default match is underlimit - so over the limit, we need to invert */ 699 return cfg->mode & XT_HASHLIMIT_INVERT; 700 701 hotdrop: 702 par->hotdrop = true; 703 return false; 704 } 705 706 static bool 707 hashlimit_mt_v1(const struct sk_buff *skb, struct xt_action_param *par) 708 { 709 const struct xt_hashlimit_mtinfo1 *info = par->matchinfo; 710 struct xt_hashlimit_htable *hinfo = info->hinfo; 711 struct hashlimit_cfg2 cfg = {}; 712 int ret; 713 714 ret = cfg_copy(&cfg, (void *)&info->cfg, 1); 715 716 if (ret) 717 return ret; 718 719 return hashlimit_mt_common(skb, par, hinfo, &cfg, 1); 720 } 721 722 static bool 723 hashlimit_mt(const struct sk_buff *skb, struct xt_action_param *par) 724 { 725 const struct xt_hashlimit_mtinfo2 *info = par->matchinfo; 726 struct xt_hashlimit_htable *hinfo = info->hinfo; 727 728 return hashlimit_mt_common(skb, par, hinfo, &info->cfg, 2); 729 } 730 731 static int hashlimit_mt_check_common(const struct xt_mtchk_param *par, 732 struct xt_hashlimit_htable **hinfo, 733 struct hashlimit_cfg2 *cfg, 734 const char *name, int revision) 735 { 736 struct net *net = par->net; 737 int ret; 738 739 if (cfg->gc_interval == 0 || cfg->expire == 0) 740 return -EINVAL; 741 if (par->family == NFPROTO_IPV4) { 742 if (cfg->srcmask > 32 || cfg->dstmask > 32) 743 return -EINVAL; 744 } else { 745 if (cfg->srcmask > 128 || cfg->dstmask > 128) 746 return -EINVAL; 747 } 748 749 if (cfg->mode & ~XT_HASHLIMIT_ALL) { 750 pr_info("Unknown mode mask %X, kernel too old?\n", 751 cfg->mode); 752 return -EINVAL; 753 } 754 755 /* Check for overflow. */ 756 if (cfg->mode & XT_HASHLIMIT_BYTES) { 757 if (user2credits_byte(cfg->avg) == 0) { 758 pr_info("overflow, rate too high: %llu\n", cfg->avg); 759 return -EINVAL; 760 } 761 } else if (cfg->burst == 0 || 762 user2credits(cfg->avg * cfg->burst, revision) < 763 user2credits(cfg->avg, revision)) { 764 pr_info("overflow, try lower: %llu/%llu\n", 765 cfg->avg, cfg->burst); 766 return -ERANGE; 767 } 768 769 mutex_lock(&hashlimit_mutex); 770 *hinfo = htable_find_get(net, name, par->family); 771 if (*hinfo == NULL) { 772 ret = htable_create(net, cfg, name, par->family, 773 hinfo, revision); 774 if (ret < 0) { 775 mutex_unlock(&hashlimit_mutex); 776 return ret; 777 } 778 } 779 mutex_unlock(&hashlimit_mutex); 780 781 return 0; 782 } 783 784 static int hashlimit_mt_check_v1(const struct xt_mtchk_param *par) 785 { 786 struct xt_hashlimit_mtinfo1 *info = par->matchinfo; 787 struct hashlimit_cfg2 cfg = {}; 788 int ret; 789 790 if (info->name[sizeof(info->name) - 1] != '\0') 791 return -EINVAL; 792 793 ret = cfg_copy(&cfg, (void *)&info->cfg, 1); 794 795 if (ret) 796 return ret; 797 798 return hashlimit_mt_check_common(par, &info->hinfo, 799 &cfg, info->name, 1); 800 } 801 802 static int hashlimit_mt_check(const struct xt_mtchk_param *par) 803 { 804 struct xt_hashlimit_mtinfo2 *info = par->matchinfo; 805 806 if (info->name[sizeof(info->name) - 1] != '\0') 807 return -EINVAL; 808 809 return hashlimit_mt_check_common(par, &info->hinfo, &info->cfg, 810 info->name, 2); 811 } 812 813 static void hashlimit_mt_destroy_v1(const struct xt_mtdtor_param *par) 814 { 815 const struct xt_hashlimit_mtinfo1 *info = par->matchinfo; 816 817 htable_put(info->hinfo); 818 } 819 820 static void hashlimit_mt_destroy(const struct xt_mtdtor_param *par) 821 { 822 const struct xt_hashlimit_mtinfo2 *info = par->matchinfo; 823 824 htable_put(info->hinfo); 825 } 826 827 static struct xt_match hashlimit_mt_reg[] __read_mostly = { 828 { 829 .name = "hashlimit", 830 .revision = 1, 831 .family = NFPROTO_IPV4, 832 .match = hashlimit_mt_v1, 833 .matchsize = sizeof(struct xt_hashlimit_mtinfo1), 834 .usersize = offsetof(struct xt_hashlimit_mtinfo1, hinfo), 835 .checkentry = hashlimit_mt_check_v1, 836 .destroy = hashlimit_mt_destroy_v1, 837 .me = THIS_MODULE, 838 }, 839 { 840 .name = "hashlimit", 841 .revision = 2, 842 .family = NFPROTO_IPV4, 843 .match = hashlimit_mt, 844 .matchsize = sizeof(struct xt_hashlimit_mtinfo2), 845 .usersize = offsetof(struct xt_hashlimit_mtinfo2, hinfo), 846 .checkentry = hashlimit_mt_check, 847 .destroy = hashlimit_mt_destroy, 848 .me = THIS_MODULE, 849 }, 850 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) 851 { 852 .name = "hashlimit", 853 .revision = 1, 854 .family = NFPROTO_IPV6, 855 .match = hashlimit_mt_v1, 856 .matchsize = sizeof(struct xt_hashlimit_mtinfo1), 857 .usersize = offsetof(struct xt_hashlimit_mtinfo1, hinfo), 858 .checkentry = hashlimit_mt_check_v1, 859 .destroy = hashlimit_mt_destroy_v1, 860 .me = THIS_MODULE, 861 }, 862 { 863 .name = "hashlimit", 864 .revision = 2, 865 .family = NFPROTO_IPV6, 866 .match = hashlimit_mt, 867 .matchsize = sizeof(struct xt_hashlimit_mtinfo2), 868 .usersize = offsetof(struct xt_hashlimit_mtinfo2, hinfo), 869 .checkentry = hashlimit_mt_check, 870 .destroy = hashlimit_mt_destroy, 871 .me = THIS_MODULE, 872 }, 873 #endif 874 }; 875 876 /* PROC stuff */ 877 static void *dl_seq_start(struct seq_file *s, loff_t *pos) 878 __acquires(htable->lock) 879 { 880 struct xt_hashlimit_htable *htable = s->private; 881 unsigned int *bucket; 882 883 spin_lock_bh(&htable->lock); 884 if (*pos >= htable->cfg.size) 885 return NULL; 886 887 bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC); 888 if (!bucket) 889 return ERR_PTR(-ENOMEM); 890 891 *bucket = *pos; 892 return bucket; 893 } 894 895 static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos) 896 { 897 struct xt_hashlimit_htable *htable = s->private; 898 unsigned int *bucket = v; 899 900 *pos = ++(*bucket); 901 if (*pos >= htable->cfg.size) { 902 kfree(v); 903 return NULL; 904 } 905 return bucket; 906 } 907 908 static void dl_seq_stop(struct seq_file *s, void *v) 909 __releases(htable->lock) 910 { 911 struct xt_hashlimit_htable *htable = s->private; 912 unsigned int *bucket = v; 913 914 if (!IS_ERR(bucket)) 915 kfree(bucket); 916 spin_unlock_bh(&htable->lock); 917 } 918 919 static void dl_seq_print(struct dsthash_ent *ent, u_int8_t family, 920 struct seq_file *s) 921 { 922 switch (family) { 923 case NFPROTO_IPV4: 924 seq_printf(s, "%ld %pI4:%u->%pI4:%u %llu %llu %llu\n", 925 (long)(ent->expires - jiffies)/HZ, 926 &ent->dst.ip.src, 927 ntohs(ent->dst.src_port), 928 &ent->dst.ip.dst, 929 ntohs(ent->dst.dst_port), 930 ent->rateinfo.credit, ent->rateinfo.credit_cap, 931 ent->rateinfo.cost); 932 break; 933 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) 934 case NFPROTO_IPV6: 935 seq_printf(s, "%ld %pI6:%u->%pI6:%u %llu %llu %llu\n", 936 (long)(ent->expires - jiffies)/HZ, 937 &ent->dst.ip6.src, 938 ntohs(ent->dst.src_port), 939 &ent->dst.ip6.dst, 940 ntohs(ent->dst.dst_port), 941 ent->rateinfo.credit, ent->rateinfo.credit_cap, 942 ent->rateinfo.cost); 943 break; 944 #endif 945 default: 946 BUG(); 947 } 948 } 949 950 static int dl_seq_real_show_v1(struct dsthash_ent *ent, u_int8_t family, 951 struct seq_file *s) 952 { 953 const struct xt_hashlimit_htable *ht = s->private; 954 955 spin_lock(&ent->lock); 956 /* recalculate to show accurate numbers */ 957 rateinfo_recalc(ent, jiffies, ht->cfg.mode, 1); 958 959 dl_seq_print(ent, family, s); 960 961 spin_unlock(&ent->lock); 962 return seq_has_overflowed(s); 963 } 964 965 static int dl_seq_real_show(struct dsthash_ent *ent, u_int8_t family, 966 struct seq_file *s) 967 { 968 const struct xt_hashlimit_htable *ht = s->private; 969 970 spin_lock(&ent->lock); 971 /* recalculate to show accurate numbers */ 972 rateinfo_recalc(ent, jiffies, ht->cfg.mode, 2); 973 974 dl_seq_print(ent, family, s); 975 976 spin_unlock(&ent->lock); 977 return seq_has_overflowed(s); 978 } 979 980 static int dl_seq_show_v1(struct seq_file *s, void *v) 981 { 982 struct xt_hashlimit_htable *htable = s->private; 983 unsigned int *bucket = v; 984 struct dsthash_ent *ent; 985 986 if (!hlist_empty(&htable->hash[*bucket])) { 987 hlist_for_each_entry(ent, &htable->hash[*bucket], node) 988 if (dl_seq_real_show_v1(ent, htable->family, s)) 989 return -1; 990 } 991 return 0; 992 } 993 994 static int dl_seq_show(struct seq_file *s, void *v) 995 { 996 struct xt_hashlimit_htable *htable = s->private; 997 unsigned int *bucket = v; 998 struct dsthash_ent *ent; 999 1000 if (!hlist_empty(&htable->hash[*bucket])) { 1001 hlist_for_each_entry(ent, &htable->hash[*bucket], node) 1002 if (dl_seq_real_show(ent, htable->family, s)) 1003 return -1; 1004 } 1005 return 0; 1006 } 1007 1008 static const struct seq_operations dl_seq_ops_v1 = { 1009 .start = dl_seq_start, 1010 .next = dl_seq_next, 1011 .stop = dl_seq_stop, 1012 .show = dl_seq_show_v1 1013 }; 1014 1015 static const struct seq_operations dl_seq_ops = { 1016 .start = dl_seq_start, 1017 .next = dl_seq_next, 1018 .stop = dl_seq_stop, 1019 .show = dl_seq_show 1020 }; 1021 1022 static int dl_proc_open_v1(struct inode *inode, struct file *file) 1023 { 1024 int ret = seq_open(file, &dl_seq_ops_v1); 1025 1026 if (!ret) { 1027 struct seq_file *sf = file->private_data; 1028 sf->private = PDE_DATA(inode); 1029 } 1030 return ret; 1031 } 1032 1033 static int dl_proc_open(struct inode *inode, struct file *file) 1034 { 1035 int ret = seq_open(file, &dl_seq_ops); 1036 1037 if (!ret) { 1038 struct seq_file *sf = file->private_data; 1039 1040 sf->private = PDE_DATA(inode); 1041 } 1042 return ret; 1043 } 1044 1045 static const struct file_operations dl_file_ops_v1 = { 1046 .owner = THIS_MODULE, 1047 .open = dl_proc_open_v1, 1048 .read = seq_read, 1049 .llseek = seq_lseek, 1050 .release = seq_release 1051 }; 1052 1053 static const struct file_operations dl_file_ops = { 1054 .owner = THIS_MODULE, 1055 .open = dl_proc_open, 1056 .read = seq_read, 1057 .llseek = seq_lseek, 1058 .release = seq_release 1059 }; 1060 1061 static int __net_init hashlimit_proc_net_init(struct net *net) 1062 { 1063 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net); 1064 1065 hashlimit_net->ipt_hashlimit = proc_mkdir("ipt_hashlimit", net->proc_net); 1066 if (!hashlimit_net->ipt_hashlimit) 1067 return -ENOMEM; 1068 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) 1069 hashlimit_net->ip6t_hashlimit = proc_mkdir("ip6t_hashlimit", net->proc_net); 1070 if (!hashlimit_net->ip6t_hashlimit) { 1071 remove_proc_entry("ipt_hashlimit", net->proc_net); 1072 return -ENOMEM; 1073 } 1074 #endif 1075 return 0; 1076 } 1077 1078 static void __net_exit hashlimit_proc_net_exit(struct net *net) 1079 { 1080 struct xt_hashlimit_htable *hinfo; 1081 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net); 1082 1083 /* hashlimit_net_exit() is called before hashlimit_mt_destroy(). 1084 * Make sure that the parent ipt_hashlimit and ip6t_hashlimit proc 1085 * entries is empty before trying to remove it. 1086 */ 1087 mutex_lock(&hashlimit_mutex); 1088 hlist_for_each_entry(hinfo, &hashlimit_net->htables, node) 1089 htable_remove_proc_entry(hinfo); 1090 hashlimit_net->ipt_hashlimit = NULL; 1091 hashlimit_net->ip6t_hashlimit = NULL; 1092 mutex_unlock(&hashlimit_mutex); 1093 1094 remove_proc_entry("ipt_hashlimit", net->proc_net); 1095 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) 1096 remove_proc_entry("ip6t_hashlimit", net->proc_net); 1097 #endif 1098 } 1099 1100 static int __net_init hashlimit_net_init(struct net *net) 1101 { 1102 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net); 1103 1104 INIT_HLIST_HEAD(&hashlimit_net->htables); 1105 return hashlimit_proc_net_init(net); 1106 } 1107 1108 static void __net_exit hashlimit_net_exit(struct net *net) 1109 { 1110 hashlimit_proc_net_exit(net); 1111 } 1112 1113 static struct pernet_operations hashlimit_net_ops = { 1114 .init = hashlimit_net_init, 1115 .exit = hashlimit_net_exit, 1116 .id = &hashlimit_net_id, 1117 .size = sizeof(struct hashlimit_net), 1118 }; 1119 1120 static int __init hashlimit_mt_init(void) 1121 { 1122 int err; 1123 1124 err = register_pernet_subsys(&hashlimit_net_ops); 1125 if (err < 0) 1126 return err; 1127 err = xt_register_matches(hashlimit_mt_reg, 1128 ARRAY_SIZE(hashlimit_mt_reg)); 1129 if (err < 0) 1130 goto err1; 1131 1132 err = -ENOMEM; 1133 hashlimit_cachep = kmem_cache_create("xt_hashlimit", 1134 sizeof(struct dsthash_ent), 0, 0, 1135 NULL); 1136 if (!hashlimit_cachep) { 1137 pr_warn("unable to create slab cache\n"); 1138 goto err2; 1139 } 1140 return 0; 1141 1142 err2: 1143 xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg)); 1144 err1: 1145 unregister_pernet_subsys(&hashlimit_net_ops); 1146 return err; 1147 1148 } 1149 1150 static void __exit hashlimit_mt_exit(void) 1151 { 1152 xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg)); 1153 unregister_pernet_subsys(&hashlimit_net_ops); 1154 1155 rcu_barrier_bh(); 1156 kmem_cache_destroy(hashlimit_cachep); 1157 } 1158 1159 module_init(hashlimit_mt_init); 1160 module_exit(hashlimit_mt_exit); 1161