| 1 | // SPDX-License-Identifier: GPL-2.0-only |
| 2 | /* |
| 3 | * xt_hashlimit - Netfilter module to limit the number of packets per time |
| 4 | * separately for each hashbucket (sourceip/sourceport/dstip/dstport) |
| 5 | * |
| 6 | * (C) 2003-2004 by Harald Welte <laforge@netfilter.org> |
| 7 | * (C) 2006-2012 Patrick McHardy <kaber@trash.net> |
| 8 | * Copyright © CC Computer Consultants GmbH, 2007 - 2008 |
| 9 | * |
| 10 | * Development of this code was funded by Astaro AG, http://www.astaro.com/ |
| 11 | */ |
| 12 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
| 13 | #include <linux/module.h> |
| 14 | #include <linux/spinlock.h> |
| 15 | #include <linux/random.h> |
| 16 | #include <linux/jhash.h> |
| 17 | #include <linux/slab.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/mutex.h> |
| 37 | #include <linux/kernel.h> |
| 38 | #include <linux/refcount.h> |
| 39 | #include <uapi/linux/netfilter/xt_hashlimit.h> |
| 40 | |
| 41 | #define XT_HASHLIMIT_ALL (XT_HASHLIMIT_HASH_DIP | XT_HASHLIMIT_HASH_DPT | \ |
| 42 | XT_HASHLIMIT_HASH_SIP | XT_HASHLIMIT_HASH_SPT | \ |
| 43 | XT_HASHLIMIT_INVERT | XT_HASHLIMIT_BYTES |\ |
| 44 | XT_HASHLIMIT_RATE_MATCH) |
| 45 | |
| 46 | MODULE_LICENSE("GPL" ); |
| 47 | MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>" ); |
| 48 | MODULE_AUTHOR("Jan Engelhardt <jengelh@medozas.de>" ); |
| 49 | MODULE_DESCRIPTION("Xtables: per hash-bucket rate-limit match" ); |
| 50 | MODULE_ALIAS("ipt_hashlimit" ); |
| 51 | MODULE_ALIAS("ip6t_hashlimit" ); |
| 52 | |
| 53 | struct hashlimit_net { |
| 54 | struct hlist_head htables; |
| 55 | struct proc_dir_entry *ipt_hashlimit; |
| 56 | struct proc_dir_entry *ip6t_hashlimit; |
| 57 | }; |
| 58 | |
| 59 | static unsigned int hashlimit_net_id; |
| 60 | static inline struct hashlimit_net *hashlimit_pernet(struct net *net) |
| 61 | { |
| 62 | return net_generic(net, id: hashlimit_net_id); |
| 63 | } |
| 64 | |
| 65 | /* need to declare this at the top */ |
| 66 | static const struct seq_operations dl_seq_ops_v2; |
| 67 | static const struct seq_operations dl_seq_ops_v1; |
| 68 | static const struct seq_operations dl_seq_ops; |
| 69 | |
| 70 | /* hash table crap */ |
| 71 | struct dsthash_dst { |
| 72 | union { |
| 73 | struct { |
| 74 | __be32 src; |
| 75 | __be32 dst; |
| 76 | } ip; |
| 77 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) |
| 78 | struct { |
| 79 | __be32 src[4]; |
| 80 | __be32 dst[4]; |
| 81 | } ip6; |
| 82 | #endif |
| 83 | }; |
| 84 | __be16 src_port; |
| 85 | __be16 dst_port; |
| 86 | }; |
| 87 | |
| 88 | struct dsthash_ent { |
| 89 | /* static / read-only parts in the beginning */ |
| 90 | struct hlist_node node; |
| 91 | struct dsthash_dst dst; |
| 92 | |
| 93 | /* modified structure members in the end */ |
| 94 | spinlock_t lock; |
| 95 | unsigned long expires; /* precalculated expiry time */ |
| 96 | struct { |
| 97 | unsigned long prev; /* last modification */ |
| 98 | union { |
| 99 | struct { |
| 100 | u_int64_t credit; |
| 101 | u_int64_t credit_cap; |
| 102 | u_int64_t cost; |
| 103 | }; |
| 104 | struct { |
| 105 | u_int32_t interval, prev_window; |
| 106 | u_int64_t current_rate; |
| 107 | u_int64_t rate; |
| 108 | int64_t burst; |
| 109 | }; |
| 110 | }; |
| 111 | } rateinfo; |
| 112 | struct rcu_head rcu; |
| 113 | }; |
| 114 | |
| 115 | struct xt_hashlimit_htable { |
| 116 | struct hlist_node node; /* global list of all htables */ |
| 117 | refcount_t use; |
| 118 | u_int8_t family; |
| 119 | bool rnd_initialized; |
| 120 | |
| 121 | struct hashlimit_cfg3 cfg; /* config */ |
| 122 | |
| 123 | /* used internally */ |
| 124 | spinlock_t lock; /* lock for list_head */ |
| 125 | u_int32_t rnd; /* random seed for hash */ |
| 126 | unsigned int count; /* number entries in table */ |
| 127 | struct delayed_work gc_work; |
| 128 | |
| 129 | /* seq_file stuff */ |
| 130 | struct proc_dir_entry *pde; |
| 131 | const char *name; |
| 132 | struct net *net; |
| 133 | |
| 134 | struct hlist_head hash[]; /* hashtable itself */ |
| 135 | }; |
| 136 | |
| 137 | static int |
| 138 | cfg_copy(struct hashlimit_cfg3 *to, const void *from, int revision) |
| 139 | { |
| 140 | if (revision == 1) { |
| 141 | struct hashlimit_cfg1 *cfg = (struct hashlimit_cfg1 *)from; |
| 142 | |
| 143 | to->mode = cfg->mode; |
| 144 | to->avg = cfg->avg; |
| 145 | to->burst = cfg->burst; |
| 146 | to->size = cfg->size; |
| 147 | to->max = cfg->max; |
| 148 | to->gc_interval = cfg->gc_interval; |
| 149 | to->expire = cfg->expire; |
| 150 | to->srcmask = cfg->srcmask; |
| 151 | to->dstmask = cfg->dstmask; |
| 152 | } else if (revision == 2) { |
| 153 | struct hashlimit_cfg2 *cfg = (struct hashlimit_cfg2 *)from; |
| 154 | |
| 155 | to->mode = cfg->mode; |
| 156 | to->avg = cfg->avg; |
| 157 | to->burst = cfg->burst; |
| 158 | to->size = cfg->size; |
| 159 | to->max = cfg->max; |
| 160 | to->gc_interval = cfg->gc_interval; |
| 161 | to->expire = cfg->expire; |
| 162 | to->srcmask = cfg->srcmask; |
| 163 | to->dstmask = cfg->dstmask; |
| 164 | } else if (revision == 3) { |
| 165 | memcpy(to, from, sizeof(struct hashlimit_cfg3)); |
| 166 | } else { |
| 167 | return -EINVAL; |
| 168 | } |
| 169 | |
| 170 | return 0; |
| 171 | } |
| 172 | |
| 173 | static DEFINE_MUTEX(hashlimit_mutex); /* protects htables list */ |
| 174 | static struct kmem_cache *hashlimit_cachep __read_mostly; |
| 175 | |
| 176 | static inline bool dst_cmp(const struct dsthash_ent *ent, |
| 177 | const struct dsthash_dst *b) |
| 178 | { |
| 179 | return !memcmp(p: &ent->dst, q: b, size: sizeof(ent->dst)); |
| 180 | } |
| 181 | |
| 182 | static u_int32_t |
| 183 | hash_dst(const struct xt_hashlimit_htable *ht, const struct dsthash_dst *dst) |
| 184 | { |
| 185 | u_int32_t hash = jhash2(k: (const u32 *)dst, |
| 186 | length: sizeof(*dst)/sizeof(u32), |
| 187 | initval: ht->rnd); |
| 188 | /* |
| 189 | * Instead of returning hash % ht->cfg.size (implying a divide) |
| 190 | * we return the high 32 bits of the (hash * ht->cfg.size) that will |
| 191 | * give results between [0 and cfg.size-1] and same hash distribution, |
| 192 | * but using a multiply, less expensive than a divide |
| 193 | */ |
| 194 | return reciprocal_scale(val: hash, ep_ro: ht->cfg.size); |
| 195 | } |
| 196 | |
| 197 | static struct dsthash_ent * |
| 198 | dsthash_find(const struct xt_hashlimit_htable *ht, |
| 199 | const struct dsthash_dst *dst) |
| 200 | { |
| 201 | struct dsthash_ent *ent; |
| 202 | u_int32_t hash = hash_dst(ht, dst); |
| 203 | |
| 204 | if (!hlist_empty(h: &ht->hash[hash])) { |
| 205 | hlist_for_each_entry_rcu(ent, &ht->hash[hash], node) |
| 206 | if (dst_cmp(ent, b: dst)) { |
| 207 | spin_lock(lock: &ent->lock); |
| 208 | return ent; |
| 209 | } |
| 210 | } |
| 211 | return NULL; |
| 212 | } |
| 213 | |
| 214 | /* allocate dsthash_ent, initialize dst, put in htable and lock it */ |
| 215 | static struct dsthash_ent * |
| 216 | dsthash_alloc_init(struct xt_hashlimit_htable *ht, |
| 217 | const struct dsthash_dst *dst, bool *race) |
| 218 | { |
| 219 | struct dsthash_ent *ent; |
| 220 | |
| 221 | spin_lock(lock: &ht->lock); |
| 222 | |
| 223 | /* Two or more packets may race to create the same entry in the |
| 224 | * hashtable, double check if this packet lost race. |
| 225 | */ |
| 226 | ent = dsthash_find(ht, dst); |
| 227 | if (ent != NULL) { |
| 228 | spin_unlock(lock: &ht->lock); |
| 229 | *race = true; |
| 230 | return ent; |
| 231 | } |
| 232 | |
| 233 | /* initialize hash with random val at the time we allocate |
| 234 | * the first hashtable entry */ |
| 235 | if (unlikely(!ht->rnd_initialized)) { |
| 236 | get_random_bytes(buf: &ht->rnd, len: sizeof(ht->rnd)); |
| 237 | ht->rnd_initialized = true; |
| 238 | } |
| 239 | |
| 240 | if (ht->cfg.max && ht->count >= ht->cfg.max) { |
| 241 | /* FIXME: do something. question is what.. */ |
| 242 | net_err_ratelimited("max count of %u reached\n" , ht->cfg.max); |
| 243 | ent = NULL; |
| 244 | } else |
| 245 | ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC); |
| 246 | if (ent) { |
| 247 | memcpy(&ent->dst, dst, sizeof(ent->dst)); |
| 248 | spin_lock_init(&ent->lock); |
| 249 | |
| 250 | spin_lock(lock: &ent->lock); |
| 251 | hlist_add_head_rcu(n: &ent->node, h: &ht->hash[hash_dst(ht, dst)]); |
| 252 | ht->count++; |
| 253 | } |
| 254 | spin_unlock(lock: &ht->lock); |
| 255 | return ent; |
| 256 | } |
| 257 | |
| 258 | static void dsthash_free_rcu(struct rcu_head *head) |
| 259 | { |
| 260 | struct dsthash_ent *ent = container_of(head, struct dsthash_ent, rcu); |
| 261 | |
| 262 | kmem_cache_free(s: hashlimit_cachep, objp: ent); |
| 263 | } |
| 264 | |
| 265 | static inline void |
| 266 | dsthash_free(struct xt_hashlimit_htable *ht, struct dsthash_ent *ent) |
| 267 | { |
| 268 | hlist_del_rcu(n: &ent->node); |
| 269 | call_rcu(head: &ent->rcu, func: dsthash_free_rcu); |
| 270 | ht->count--; |
| 271 | } |
| 272 | static void htable_gc(struct work_struct *work); |
| 273 | |
| 274 | static int htable_create(struct net *net, struct hashlimit_cfg3 *cfg, |
| 275 | const char *name, u_int8_t family, |
| 276 | struct xt_hashlimit_htable **out_hinfo, |
| 277 | int revision) |
| 278 | { |
| 279 | struct hashlimit_net *hashlimit_net = hashlimit_pernet(net); |
| 280 | struct xt_hashlimit_htable *hinfo; |
| 281 | const struct seq_operations *ops; |
| 282 | unsigned int size, i; |
| 283 | unsigned long nr_pages = totalram_pages(); |
| 284 | int ret; |
| 285 | |
| 286 | if (cfg->size) { |
| 287 | size = cfg->size; |
| 288 | } else { |
| 289 | size = (nr_pages << PAGE_SHIFT) / 16384 / |
| 290 | sizeof(struct hlist_head); |
| 291 | if (nr_pages > 1024 * 1024 * 1024 / PAGE_SIZE) |
| 292 | size = 8192; |
| 293 | if (size < 16) |
| 294 | size = 16; |
| 295 | } |
| 296 | hinfo = kvmalloc(struct_size(hinfo, hash, size), GFP_KERNEL); |
| 297 | if (hinfo == NULL) |
| 298 | return -ENOMEM; |
| 299 | *out_hinfo = hinfo; |
| 300 | |
| 301 | /* copy match config into hashtable config */ |
| 302 | ret = cfg_copy(to: &hinfo->cfg, from: (void *)cfg, revision: 3); |
| 303 | if (ret) { |
| 304 | kvfree(addr: hinfo); |
| 305 | return ret; |
| 306 | } |
| 307 | |
| 308 | hinfo->cfg.size = size; |
| 309 | if (hinfo->cfg.max == 0) |
| 310 | hinfo->cfg.max = 8 * hinfo->cfg.size; |
| 311 | else if (hinfo->cfg.max < hinfo->cfg.size) |
| 312 | hinfo->cfg.max = hinfo->cfg.size; |
| 313 | |
| 314 | for (i = 0; i < hinfo->cfg.size; i++) |
| 315 | INIT_HLIST_HEAD(&hinfo->hash[i]); |
| 316 | |
| 317 | refcount_set(r: &hinfo->use, n: 1); |
| 318 | hinfo->count = 0; |
| 319 | hinfo->family = family; |
| 320 | hinfo->rnd_initialized = false; |
| 321 | hinfo->name = kstrdup(s: name, GFP_KERNEL); |
| 322 | if (!hinfo->name) { |
| 323 | kvfree(addr: hinfo); |
| 324 | return -ENOMEM; |
| 325 | } |
| 326 | spin_lock_init(&hinfo->lock); |
| 327 | |
| 328 | switch (revision) { |
| 329 | case 1: |
| 330 | ops = &dl_seq_ops_v1; |
| 331 | break; |
| 332 | case 2: |
| 333 | ops = &dl_seq_ops_v2; |
| 334 | break; |
| 335 | default: |
| 336 | ops = &dl_seq_ops; |
| 337 | } |
| 338 | |
| 339 | hinfo->pde = proc_create_seq_data(name, 0, |
| 340 | (family == NFPROTO_IPV4) ? |
| 341 | hashlimit_net->ipt_hashlimit : hashlimit_net->ip6t_hashlimit, |
| 342 | ops, hinfo); |
| 343 | if (hinfo->pde == NULL) { |
| 344 | kfree(objp: hinfo->name); |
| 345 | kvfree(addr: hinfo); |
| 346 | return -ENOMEM; |
| 347 | } |
| 348 | hinfo->net = net; |
| 349 | |
| 350 | INIT_DEFERRABLE_WORK(&hinfo->gc_work, htable_gc); |
| 351 | queue_delayed_work(wq: system_power_efficient_wq, dwork: &hinfo->gc_work, |
| 352 | delay: msecs_to_jiffies(m: hinfo->cfg.gc_interval)); |
| 353 | |
| 354 | hlist_add_head(n: &hinfo->node, h: &hashlimit_net->htables); |
| 355 | |
| 356 | return 0; |
| 357 | } |
| 358 | |
| 359 | static void htable_selective_cleanup(struct xt_hashlimit_htable *ht, bool select_all) |
| 360 | { |
| 361 | unsigned int i; |
| 362 | |
| 363 | for (i = 0; i < ht->cfg.size; i++) { |
| 364 | struct hlist_head *head = &ht->hash[i]; |
| 365 | struct dsthash_ent *dh; |
| 366 | struct hlist_node *n; |
| 367 | |
| 368 | if (hlist_empty(h: head)) |
| 369 | continue; |
| 370 | |
| 371 | spin_lock_bh(lock: &ht->lock); |
| 372 | hlist_for_each_entry_safe(dh, n, head, node) { |
| 373 | if (time_after_eq(jiffies, dh->expires) || select_all) |
| 374 | dsthash_free(ht, ent: dh); |
| 375 | } |
| 376 | spin_unlock_bh(lock: &ht->lock); |
| 377 | cond_resched(); |
| 378 | } |
| 379 | } |
| 380 | |
| 381 | static void htable_gc(struct work_struct *work) |
| 382 | { |
| 383 | struct xt_hashlimit_htable *ht; |
| 384 | |
| 385 | ht = container_of(work, struct xt_hashlimit_htable, gc_work.work); |
| 386 | |
| 387 | htable_selective_cleanup(ht, select_all: false); |
| 388 | |
| 389 | queue_delayed_work(wq: system_power_efficient_wq, |
| 390 | dwork: &ht->gc_work, delay: msecs_to_jiffies(m: ht->cfg.gc_interval)); |
| 391 | } |
| 392 | |
| 393 | static void htable_remove_proc_entry(struct xt_hashlimit_htable *hinfo) |
| 394 | { |
| 395 | struct hashlimit_net *hashlimit_net = hashlimit_pernet(net: hinfo->net); |
| 396 | struct proc_dir_entry *parent; |
| 397 | |
| 398 | if (hinfo->family == NFPROTO_IPV4) |
| 399 | parent = hashlimit_net->ipt_hashlimit; |
| 400 | else |
| 401 | parent = hashlimit_net->ip6t_hashlimit; |
| 402 | |
| 403 | if (parent != NULL) |
| 404 | remove_proc_entry(hinfo->name, parent); |
| 405 | } |
| 406 | |
| 407 | static struct xt_hashlimit_htable *htable_find_get(struct net *net, |
| 408 | const char *name, |
| 409 | u_int8_t family) |
| 410 | { |
| 411 | struct hashlimit_net *hashlimit_net = hashlimit_pernet(net); |
| 412 | struct xt_hashlimit_htable *hinfo; |
| 413 | |
| 414 | hlist_for_each_entry(hinfo, &hashlimit_net->htables, node) { |
| 415 | if (!strcmp(name, hinfo->name) && |
| 416 | hinfo->family == family) { |
| 417 | refcount_inc(r: &hinfo->use); |
| 418 | return hinfo; |
| 419 | } |
| 420 | } |
| 421 | return NULL; |
| 422 | } |
| 423 | |
| 424 | static void htable_put(struct xt_hashlimit_htable *hinfo) |
| 425 | { |
| 426 | if (refcount_dec_and_mutex_lock(r: &hinfo->use, lock: &hashlimit_mutex)) { |
| 427 | hlist_del(n: &hinfo->node); |
| 428 | htable_remove_proc_entry(hinfo); |
| 429 | mutex_unlock(lock: &hashlimit_mutex); |
| 430 | |
| 431 | cancel_delayed_work_sync(dwork: &hinfo->gc_work); |
| 432 | htable_selective_cleanup(ht: hinfo, select_all: true); |
| 433 | kfree(objp: hinfo->name); |
| 434 | kvfree(addr: hinfo); |
| 435 | } |
| 436 | } |
| 437 | |
| 438 | /* The algorithm used is the Simple Token Bucket Filter (TBF) |
| 439 | * see net/sched/sch_tbf.c in the linux source tree |
| 440 | */ |
| 441 | |
| 442 | /* Rusty: This is my (non-mathematically-inclined) understanding of |
| 443 | this algorithm. The `average rate' in jiffies becomes your initial |
| 444 | amount of credit `credit' and the most credit you can ever have |
| 445 | `credit_cap'. The `peak rate' becomes the cost of passing the |
| 446 | test, `cost'. |
| 447 | |
| 448 | `prev' tracks the last packet hit: you gain one credit per jiffy. |
| 449 | If you get credit balance more than this, the extra credit is |
| 450 | discarded. Every time the match passes, you lose `cost' credits; |
| 451 | if you don't have that many, the test fails. |
| 452 | |
| 453 | See Alexey's formal explanation in net/sched/sch_tbf.c. |
| 454 | |
| 455 | To get the maximum range, we multiply by this factor (ie. you get N |
| 456 | credits per jiffy). We want to allow a rate as low as 1 per day |
| 457 | (slowest userspace tool allows), which means |
| 458 | CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie. |
| 459 | */ |
| 460 | #define MAX_CPJ_v1 (0xFFFFFFFF / (HZ*60*60*24)) |
| 461 | #define MAX_CPJ (0xFFFFFFFFFFFFFFFFULL / (HZ*60*60*24)) |
| 462 | |
| 463 | /* Repeated shift and or gives us all 1s, final shift and add 1 gives |
| 464 | * us the power of 2 below the theoretical max, so GCC simply does a |
| 465 | * shift. */ |
| 466 | #define _POW2_BELOW2(x) ((x)|((x)>>1)) |
| 467 | #define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2)) |
| 468 | #define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4)) |
| 469 | #define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8)) |
| 470 | #define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16)) |
| 471 | #define _POW2_BELOW64(x) (_POW2_BELOW32(x)|_POW2_BELOW32((x)>>32)) |
| 472 | #define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1) |
| 473 | #define POW2_BELOW64(x) ((_POW2_BELOW64(x)>>1) + 1) |
| 474 | |
| 475 | #define CREDITS_PER_JIFFY POW2_BELOW64(MAX_CPJ) |
| 476 | #define CREDITS_PER_JIFFY_v1 POW2_BELOW32(MAX_CPJ_v1) |
| 477 | |
| 478 | /* in byte mode, the lowest possible rate is one packet/second. |
| 479 | * credit_cap is used as a counter that tells us how many times we can |
| 480 | * refill the "credits available" counter when it becomes empty. |
| 481 | */ |
| 482 | #define MAX_CPJ_BYTES (0xFFFFFFFF / HZ) |
| 483 | #define CREDITS_PER_JIFFY_BYTES POW2_BELOW32(MAX_CPJ_BYTES) |
| 484 | |
| 485 | static u32 xt_hashlimit_len_to_chunks(u32 len) |
| 486 | { |
| 487 | return (len >> XT_HASHLIMIT_BYTE_SHIFT) + 1; |
| 488 | } |
| 489 | |
| 490 | /* Precision saver. */ |
| 491 | static u64 user2credits(u64 user, int revision) |
| 492 | { |
| 493 | u64 scale = (revision == 1) ? |
| 494 | XT_HASHLIMIT_SCALE : XT_HASHLIMIT_SCALE_v2; |
| 495 | u64 cpj = (revision == 1) ? |
| 496 | CREDITS_PER_JIFFY_v1 : CREDITS_PER_JIFFY; |
| 497 | |
| 498 | /* Avoid overflow: divide the constant operands first */ |
| 499 | if (scale >= HZ * cpj) |
| 500 | return div64_u64(dividend: user, divisor: div64_u64(dividend: scale, HZ * cpj)); |
| 501 | |
| 502 | return user * div64_u64(HZ * cpj, divisor: scale); |
| 503 | } |
| 504 | |
| 505 | static u32 user2credits_byte(u32 user) |
| 506 | { |
| 507 | u64 us = user; |
| 508 | us *= HZ * CREDITS_PER_JIFFY_BYTES; |
| 509 | return (u32) (us >> 32); |
| 510 | } |
| 511 | |
| 512 | static u64 user2rate(u64 user) |
| 513 | { |
| 514 | if (user != 0) { |
| 515 | return div64_u64(XT_HASHLIMIT_SCALE_v2, divisor: user); |
| 516 | } else { |
| 517 | pr_info_ratelimited("invalid rate from userspace: %llu\n" , |
| 518 | user); |
| 519 | return 0; |
| 520 | } |
| 521 | } |
| 522 | |
| 523 | static u64 user2rate_bytes(u32 user) |
| 524 | { |
| 525 | u64 r; |
| 526 | |
| 527 | r = user ? U32_MAX / user : U32_MAX; |
| 528 | return (r - 1) << XT_HASHLIMIT_BYTE_SHIFT; |
| 529 | } |
| 530 | |
| 531 | static void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now, |
| 532 | u32 mode, int revision) |
| 533 | { |
| 534 | unsigned long delta = now - dh->rateinfo.prev; |
| 535 | u64 cap, cpj; |
| 536 | |
| 537 | if (delta == 0) |
| 538 | return; |
| 539 | |
| 540 | if (revision >= 3 && mode & XT_HASHLIMIT_RATE_MATCH) { |
| 541 | u64 interval = dh->rateinfo.interval * HZ; |
| 542 | |
| 543 | if (delta < interval) |
| 544 | return; |
| 545 | |
| 546 | dh->rateinfo.prev = now; |
| 547 | dh->rateinfo.prev_window = |
| 548 | ((dh->rateinfo.current_rate * interval) > |
| 549 | (delta * dh->rateinfo.rate)); |
| 550 | dh->rateinfo.current_rate = 0; |
| 551 | |
| 552 | return; |
| 553 | } |
| 554 | |
| 555 | dh->rateinfo.prev = now; |
| 556 | |
| 557 | if (mode & XT_HASHLIMIT_BYTES) { |
| 558 | u64 tmp = dh->rateinfo.credit; |
| 559 | dh->rateinfo.credit += CREDITS_PER_JIFFY_BYTES * delta; |
| 560 | cap = CREDITS_PER_JIFFY_BYTES * HZ; |
| 561 | if (tmp >= dh->rateinfo.credit) {/* overflow */ |
| 562 | dh->rateinfo.credit = cap; |
| 563 | return; |
| 564 | } |
| 565 | } else { |
| 566 | cpj = (revision == 1) ? |
| 567 | CREDITS_PER_JIFFY_v1 : CREDITS_PER_JIFFY; |
| 568 | dh->rateinfo.credit += delta * cpj; |
| 569 | cap = dh->rateinfo.credit_cap; |
| 570 | } |
| 571 | if (dh->rateinfo.credit > cap) |
| 572 | dh->rateinfo.credit = cap; |
| 573 | } |
| 574 | |
| 575 | static void rateinfo_init(struct dsthash_ent *dh, |
| 576 | struct xt_hashlimit_htable *hinfo, int revision) |
| 577 | { |
| 578 | dh->rateinfo.prev = jiffies; |
| 579 | if (revision >= 3 && hinfo->cfg.mode & XT_HASHLIMIT_RATE_MATCH) { |
| 580 | dh->rateinfo.prev_window = 0; |
| 581 | dh->rateinfo.current_rate = 0; |
| 582 | if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) { |
| 583 | dh->rateinfo.rate = |
| 584 | user2rate_bytes(user: (u32)hinfo->cfg.avg); |
| 585 | if (hinfo->cfg.burst) |
| 586 | dh->rateinfo.burst = |
| 587 | hinfo->cfg.burst * dh->rateinfo.rate; |
| 588 | else |
| 589 | dh->rateinfo.burst = dh->rateinfo.rate; |
| 590 | } else { |
| 591 | dh->rateinfo.rate = user2rate(user: hinfo->cfg.avg); |
| 592 | dh->rateinfo.burst = |
| 593 | hinfo->cfg.burst + dh->rateinfo.rate; |
| 594 | } |
| 595 | dh->rateinfo.interval = hinfo->cfg.interval; |
| 596 | } else if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) { |
| 597 | dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ; |
| 598 | dh->rateinfo.cost = user2credits_byte(user: hinfo->cfg.avg); |
| 599 | dh->rateinfo.credit_cap = hinfo->cfg.burst; |
| 600 | } else { |
| 601 | dh->rateinfo.credit = user2credits(user: hinfo->cfg.avg * |
| 602 | hinfo->cfg.burst, revision); |
| 603 | dh->rateinfo.cost = user2credits(user: hinfo->cfg.avg, revision); |
| 604 | dh->rateinfo.credit_cap = dh->rateinfo.credit; |
| 605 | } |
| 606 | } |
| 607 | |
| 608 | static inline __be32 maskl(__be32 a, unsigned int l) |
| 609 | { |
| 610 | return l ? htonl(ntohl(a) & ~0 << (32 - l)) : 0; |
| 611 | } |
| 612 | |
| 613 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) |
| 614 | static void hashlimit_ipv6_mask(__be32 *i, unsigned int p) |
| 615 | { |
| 616 | switch (p) { |
| 617 | case 0 ... 31: |
| 618 | i[0] = maskl(a: i[0], l: p); |
| 619 | i[1] = i[2] = i[3] = 0; |
| 620 | break; |
| 621 | case 32 ... 63: |
| 622 | i[1] = maskl(a: i[1], l: p - 32); |
| 623 | i[2] = i[3] = 0; |
| 624 | break; |
| 625 | case 64 ... 95: |
| 626 | i[2] = maskl(a: i[2], l: p - 64); |
| 627 | i[3] = 0; |
| 628 | break; |
| 629 | case 96 ... 127: |
| 630 | i[3] = maskl(a: i[3], l: p - 96); |
| 631 | break; |
| 632 | case 128: |
| 633 | break; |
| 634 | } |
| 635 | } |
| 636 | #endif |
| 637 | |
| 638 | static int |
| 639 | hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo, |
| 640 | struct dsthash_dst *dst, |
| 641 | const struct sk_buff *skb, unsigned int protoff) |
| 642 | { |
| 643 | __be16 _ports[2], *ports; |
| 644 | u8 nexthdr; |
| 645 | int poff; |
| 646 | |
| 647 | memset(dst, 0, sizeof(*dst)); |
| 648 | |
| 649 | switch (hinfo->family) { |
| 650 | case NFPROTO_IPV4: |
| 651 | if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) |
| 652 | dst->ip.dst = maskl(a: ip_hdr(skb)->daddr, |
| 653 | l: hinfo->cfg.dstmask); |
| 654 | if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) |
| 655 | dst->ip.src = maskl(a: ip_hdr(skb)->saddr, |
| 656 | l: hinfo->cfg.srcmask); |
| 657 | |
| 658 | if (!(hinfo->cfg.mode & |
| 659 | (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT))) |
| 660 | return 0; |
| 661 | nexthdr = ip_hdr(skb)->protocol; |
| 662 | break; |
| 663 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) |
| 664 | case NFPROTO_IPV6: |
| 665 | { |
| 666 | __be16 frag_off; |
| 667 | |
| 668 | if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) { |
| 669 | memcpy(&dst->ip6.dst, &ipv6_hdr(skb)->daddr, |
| 670 | sizeof(dst->ip6.dst)); |
| 671 | hashlimit_ipv6_mask(i: dst->ip6.dst, p: hinfo->cfg.dstmask); |
| 672 | } |
| 673 | if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) { |
| 674 | memcpy(&dst->ip6.src, &ipv6_hdr(skb)->saddr, |
| 675 | sizeof(dst->ip6.src)); |
| 676 | hashlimit_ipv6_mask(i: dst->ip6.src, p: hinfo->cfg.srcmask); |
| 677 | } |
| 678 | |
| 679 | if (!(hinfo->cfg.mode & |
| 680 | (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT))) |
| 681 | return 0; |
| 682 | nexthdr = ipv6_hdr(skb)->nexthdr; |
| 683 | protoff = ipv6_skip_exthdr(skb, start: sizeof(struct ipv6hdr), nexthdrp: &nexthdr, frag_offp: &frag_off); |
| 684 | if ((int)protoff < 0) |
| 685 | return -1; |
| 686 | break; |
| 687 | } |
| 688 | #endif |
| 689 | default: |
| 690 | BUG(); |
| 691 | return 0; |
| 692 | } |
| 693 | |
| 694 | poff = proto_ports_offset(proto: nexthdr); |
| 695 | if (poff >= 0) { |
| 696 | ports = skb_header_pointer(skb, offset: protoff + poff, len: sizeof(_ports), |
| 697 | buffer: &_ports); |
| 698 | } else { |
| 699 | _ports[0] = _ports[1] = 0; |
| 700 | ports = _ports; |
| 701 | } |
| 702 | if (!ports) |
| 703 | return -1; |
| 704 | if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SPT) |
| 705 | dst->src_port = ports[0]; |
| 706 | if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DPT) |
| 707 | dst->dst_port = ports[1]; |
| 708 | return 0; |
| 709 | } |
| 710 | |
| 711 | static u32 hashlimit_byte_cost(unsigned int len, struct dsthash_ent *dh) |
| 712 | { |
| 713 | u64 tmp = xt_hashlimit_len_to_chunks(len); |
| 714 | tmp = tmp * dh->rateinfo.cost; |
| 715 | |
| 716 | if (unlikely(tmp > CREDITS_PER_JIFFY_BYTES * HZ)) |
| 717 | tmp = CREDITS_PER_JIFFY_BYTES * HZ; |
| 718 | |
| 719 | if (dh->rateinfo.credit < tmp && dh->rateinfo.credit_cap) { |
| 720 | dh->rateinfo.credit_cap--; |
| 721 | dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ; |
| 722 | } |
| 723 | return (u32) tmp; |
| 724 | } |
| 725 | |
| 726 | static bool |
| 727 | hashlimit_mt_common(const struct sk_buff *skb, struct xt_action_param *par, |
| 728 | struct xt_hashlimit_htable *hinfo, |
| 729 | const struct hashlimit_cfg3 *cfg, int revision) |
| 730 | { |
| 731 | unsigned long now = jiffies; |
| 732 | struct dsthash_ent *dh; |
| 733 | struct dsthash_dst dst; |
| 734 | bool race = false; |
| 735 | u64 cost; |
| 736 | |
| 737 | if (hashlimit_init_dst(hinfo, dst: &dst, skb, protoff: par->thoff) < 0) |
| 738 | goto hotdrop; |
| 739 | |
| 740 | local_bh_disable(); |
| 741 | dh = dsthash_find(ht: hinfo, dst: &dst); |
| 742 | if (dh == NULL) { |
| 743 | dh = dsthash_alloc_init(ht: hinfo, dst: &dst, race: &race); |
| 744 | if (dh == NULL) { |
| 745 | local_bh_enable(); |
| 746 | goto hotdrop; |
| 747 | } else if (race) { |
| 748 | /* Already got an entry, update expiration timeout */ |
| 749 | dh->expires = now + msecs_to_jiffies(m: hinfo->cfg.expire); |
| 750 | rateinfo_recalc(dh, now, mode: hinfo->cfg.mode, revision); |
| 751 | } else { |
| 752 | dh->expires = jiffies + msecs_to_jiffies(m: hinfo->cfg.expire); |
| 753 | rateinfo_init(dh, hinfo, revision); |
| 754 | } |
| 755 | } else { |
| 756 | /* update expiration timeout */ |
| 757 | dh->expires = now + msecs_to_jiffies(m: hinfo->cfg.expire); |
| 758 | rateinfo_recalc(dh, now, mode: hinfo->cfg.mode, revision); |
| 759 | } |
| 760 | |
| 761 | if (cfg->mode & XT_HASHLIMIT_RATE_MATCH) { |
| 762 | cost = (cfg->mode & XT_HASHLIMIT_BYTES) ? skb->len : 1; |
| 763 | dh->rateinfo.current_rate += cost; |
| 764 | |
| 765 | if (!dh->rateinfo.prev_window && |
| 766 | (dh->rateinfo.current_rate <= dh->rateinfo.burst)) { |
| 767 | spin_unlock(lock: &dh->lock); |
| 768 | local_bh_enable(); |
| 769 | return !(cfg->mode & XT_HASHLIMIT_INVERT); |
| 770 | } else { |
| 771 | goto overlimit; |
| 772 | } |
| 773 | } |
| 774 | |
| 775 | if (cfg->mode & XT_HASHLIMIT_BYTES) |
| 776 | cost = hashlimit_byte_cost(len: skb->len, dh); |
| 777 | else |
| 778 | cost = dh->rateinfo.cost; |
| 779 | |
| 780 | if (dh->rateinfo.credit >= cost) { |
| 781 | /* below the limit */ |
| 782 | dh->rateinfo.credit -= cost; |
| 783 | spin_unlock(lock: &dh->lock); |
| 784 | local_bh_enable(); |
| 785 | return !(cfg->mode & XT_HASHLIMIT_INVERT); |
| 786 | } |
| 787 | |
| 788 | overlimit: |
| 789 | spin_unlock(lock: &dh->lock); |
| 790 | local_bh_enable(); |
| 791 | /* default match is underlimit - so over the limit, we need to invert */ |
| 792 | return cfg->mode & XT_HASHLIMIT_INVERT; |
| 793 | |
| 794 | hotdrop: |
| 795 | par->hotdrop = true; |
| 796 | return false; |
| 797 | } |
| 798 | |
| 799 | static bool |
| 800 | hashlimit_mt_v1(const struct sk_buff *skb, struct xt_action_param *par) |
| 801 | { |
| 802 | const struct xt_hashlimit_mtinfo1 *info = par->matchinfo; |
| 803 | struct xt_hashlimit_htable *hinfo = info->hinfo; |
| 804 | struct hashlimit_cfg3 cfg = {}; |
| 805 | int ret; |
| 806 | |
| 807 | ret = cfg_copy(to: &cfg, from: (void *)&info->cfg, revision: 1); |
| 808 | if (ret) |
| 809 | return ret; |
| 810 | |
| 811 | return hashlimit_mt_common(skb, par, hinfo, cfg: &cfg, revision: 1); |
| 812 | } |
| 813 | |
| 814 | static bool |
| 815 | hashlimit_mt_v2(const struct sk_buff *skb, struct xt_action_param *par) |
| 816 | { |
| 817 | const struct xt_hashlimit_mtinfo2 *info = par->matchinfo; |
| 818 | struct xt_hashlimit_htable *hinfo = info->hinfo; |
| 819 | struct hashlimit_cfg3 cfg = {}; |
| 820 | int ret; |
| 821 | |
| 822 | ret = cfg_copy(to: &cfg, from: (void *)&info->cfg, revision: 2); |
| 823 | if (ret) |
| 824 | return ret; |
| 825 | |
| 826 | return hashlimit_mt_common(skb, par, hinfo, cfg: &cfg, revision: 2); |
| 827 | } |
| 828 | |
| 829 | static bool |
| 830 | hashlimit_mt(const struct sk_buff *skb, struct xt_action_param *par) |
| 831 | { |
| 832 | const struct xt_hashlimit_mtinfo3 *info = par->matchinfo; |
| 833 | struct xt_hashlimit_htable *hinfo = info->hinfo; |
| 834 | |
| 835 | return hashlimit_mt_common(skb, par, hinfo, cfg: &info->cfg, revision: 3); |
| 836 | } |
| 837 | |
| 838 | #define HASHLIMIT_MAX_SIZE 1048576 |
| 839 | |
| 840 | static int hashlimit_mt_check_common(const struct xt_mtchk_param *par, |
| 841 | struct xt_hashlimit_htable **hinfo, |
| 842 | struct hashlimit_cfg3 *cfg, |
| 843 | const char *name, int revision) |
| 844 | { |
| 845 | struct net *net = par->net; |
| 846 | int ret; |
| 847 | |
| 848 | if (cfg->gc_interval == 0 || cfg->expire == 0) |
| 849 | return -EINVAL; |
| 850 | if (cfg->size > HASHLIMIT_MAX_SIZE) { |
| 851 | cfg->size = HASHLIMIT_MAX_SIZE; |
| 852 | pr_info_ratelimited("size too large, truncated to %u\n" , cfg->size); |
| 853 | } |
| 854 | if (cfg->max > HASHLIMIT_MAX_SIZE) { |
| 855 | cfg->max = HASHLIMIT_MAX_SIZE; |
| 856 | pr_info_ratelimited("max too large, truncated to %u\n" , cfg->max); |
| 857 | } |
| 858 | if (par->family == NFPROTO_IPV4) { |
| 859 | if (cfg->srcmask > 32 || cfg->dstmask > 32) |
| 860 | return -EINVAL; |
| 861 | } else { |
| 862 | if (cfg->srcmask > 128 || cfg->dstmask > 128) |
| 863 | return -EINVAL; |
| 864 | } |
| 865 | |
| 866 | if (cfg->mode & ~XT_HASHLIMIT_ALL) { |
| 867 | pr_info_ratelimited("Unknown mode mask %X, kernel too old?\n" , |
| 868 | cfg->mode); |
| 869 | return -EINVAL; |
| 870 | } |
| 871 | |
| 872 | /* Check for overflow. */ |
| 873 | if (revision >= 3 && cfg->mode & XT_HASHLIMIT_RATE_MATCH) { |
| 874 | if (cfg->avg == 0 || cfg->avg > U32_MAX) { |
| 875 | pr_info_ratelimited("invalid rate\n" ); |
| 876 | return -ERANGE; |
| 877 | } |
| 878 | |
| 879 | if (cfg->interval == 0) { |
| 880 | pr_info_ratelimited("invalid interval\n" ); |
| 881 | return -EINVAL; |
| 882 | } |
| 883 | } else if (cfg->mode & XT_HASHLIMIT_BYTES) { |
| 884 | if (user2credits_byte(user: cfg->avg) == 0) { |
| 885 | pr_info_ratelimited("overflow, rate too high: %llu\n" , |
| 886 | cfg->avg); |
| 887 | return -EINVAL; |
| 888 | } |
| 889 | } else if (cfg->burst == 0 || |
| 890 | user2credits(user: cfg->avg * cfg->burst, revision) < |
| 891 | user2credits(user: cfg->avg, revision)) { |
| 892 | pr_info_ratelimited("overflow, try lower: %llu/%llu\n" , |
| 893 | cfg->avg, cfg->burst); |
| 894 | return -ERANGE; |
| 895 | } |
| 896 | |
| 897 | mutex_lock(&hashlimit_mutex); |
| 898 | *hinfo = htable_find_get(net, name, family: par->family); |
| 899 | if (*hinfo == NULL) { |
| 900 | ret = htable_create(net, cfg, name, family: par->family, |
| 901 | out_hinfo: hinfo, revision); |
| 902 | if (ret < 0) { |
| 903 | mutex_unlock(lock: &hashlimit_mutex); |
| 904 | return ret; |
| 905 | } |
| 906 | } |
| 907 | mutex_unlock(lock: &hashlimit_mutex); |
| 908 | |
| 909 | return 0; |
| 910 | } |
| 911 | |
| 912 | static int hashlimit_mt_check_v1(const struct xt_mtchk_param *par) |
| 913 | { |
| 914 | struct xt_hashlimit_mtinfo1 *info = par->matchinfo; |
| 915 | struct hashlimit_cfg3 cfg = {}; |
| 916 | int ret; |
| 917 | |
| 918 | ret = xt_check_proc_name(name: info->name, size: sizeof(info->name)); |
| 919 | if (ret) |
| 920 | return ret; |
| 921 | |
| 922 | ret = cfg_copy(to: &cfg, from: (void *)&info->cfg, revision: 1); |
| 923 | if (ret) |
| 924 | return ret; |
| 925 | |
| 926 | return hashlimit_mt_check_common(par, hinfo: &info->hinfo, |
| 927 | cfg: &cfg, name: info->name, revision: 1); |
| 928 | } |
| 929 | |
| 930 | static int hashlimit_mt_check_v2(const struct xt_mtchk_param *par) |
| 931 | { |
| 932 | struct xt_hashlimit_mtinfo2 *info = par->matchinfo; |
| 933 | struct hashlimit_cfg3 cfg = {}; |
| 934 | int ret; |
| 935 | |
| 936 | ret = xt_check_proc_name(name: info->name, size: sizeof(info->name)); |
| 937 | if (ret) |
| 938 | return ret; |
| 939 | |
| 940 | ret = cfg_copy(to: &cfg, from: (void *)&info->cfg, revision: 2); |
| 941 | if (ret) |
| 942 | return ret; |
| 943 | |
| 944 | return hashlimit_mt_check_common(par, hinfo: &info->hinfo, |
| 945 | cfg: &cfg, name: info->name, revision: 2); |
| 946 | } |
| 947 | |
| 948 | static int hashlimit_mt_check(const struct xt_mtchk_param *par) |
| 949 | { |
| 950 | struct xt_hashlimit_mtinfo3 *info = par->matchinfo; |
| 951 | int ret; |
| 952 | |
| 953 | ret = xt_check_proc_name(name: info->name, size: sizeof(info->name)); |
| 954 | if (ret) |
| 955 | return ret; |
| 956 | |
| 957 | return hashlimit_mt_check_common(par, hinfo: &info->hinfo, cfg: &info->cfg, |
| 958 | name: info->name, revision: 3); |
| 959 | } |
| 960 | |
| 961 | static void hashlimit_mt_destroy_v2(const struct xt_mtdtor_param *par) |
| 962 | { |
| 963 | const struct xt_hashlimit_mtinfo2 *info = par->matchinfo; |
| 964 | |
| 965 | htable_put(hinfo: info->hinfo); |
| 966 | } |
| 967 | |
| 968 | static void hashlimit_mt_destroy_v1(const struct xt_mtdtor_param *par) |
| 969 | { |
| 970 | const struct xt_hashlimit_mtinfo1 *info = par->matchinfo; |
| 971 | |
| 972 | htable_put(hinfo: info->hinfo); |
| 973 | } |
| 974 | |
| 975 | static void hashlimit_mt_destroy(const struct xt_mtdtor_param *par) |
| 976 | { |
| 977 | const struct xt_hashlimit_mtinfo3 *info = par->matchinfo; |
| 978 | |
| 979 | htable_put(hinfo: info->hinfo); |
| 980 | } |
| 981 | |
| 982 | static struct xt_match hashlimit_mt_reg[] __read_mostly = { |
| 983 | { |
| 984 | .name = "hashlimit" , |
| 985 | .revision = 1, |
| 986 | .family = NFPROTO_IPV4, |
| 987 | .match = hashlimit_mt_v1, |
| 988 | .matchsize = sizeof(struct xt_hashlimit_mtinfo1), |
| 989 | .usersize = offsetof(struct xt_hashlimit_mtinfo1, hinfo), |
| 990 | .checkentry = hashlimit_mt_check_v1, |
| 991 | .destroy = hashlimit_mt_destroy_v1, |
| 992 | .me = THIS_MODULE, |
| 993 | }, |
| 994 | { |
| 995 | .name = "hashlimit" , |
| 996 | .revision = 2, |
| 997 | .family = NFPROTO_IPV4, |
| 998 | .match = hashlimit_mt_v2, |
| 999 | .matchsize = sizeof(struct xt_hashlimit_mtinfo2), |
| 1000 | .usersize = offsetof(struct xt_hashlimit_mtinfo2, hinfo), |
| 1001 | .checkentry = hashlimit_mt_check_v2, |
| 1002 | .destroy = hashlimit_mt_destroy_v2, |
| 1003 | .me = THIS_MODULE, |
| 1004 | }, |
| 1005 | { |
| 1006 | .name = "hashlimit" , |
| 1007 | .revision = 3, |
| 1008 | .family = NFPROTO_IPV4, |
| 1009 | .match = hashlimit_mt, |
| 1010 | .matchsize = sizeof(struct xt_hashlimit_mtinfo3), |
| 1011 | .usersize = offsetof(struct xt_hashlimit_mtinfo3, hinfo), |
| 1012 | .checkentry = hashlimit_mt_check, |
| 1013 | .destroy = hashlimit_mt_destroy, |
| 1014 | .me = THIS_MODULE, |
| 1015 | }, |
| 1016 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) |
| 1017 | { |
| 1018 | .name = "hashlimit" , |
| 1019 | .revision = 1, |
| 1020 | .family = NFPROTO_IPV6, |
| 1021 | .match = hashlimit_mt_v1, |
| 1022 | .matchsize = sizeof(struct xt_hashlimit_mtinfo1), |
| 1023 | .usersize = offsetof(struct xt_hashlimit_mtinfo1, hinfo), |
| 1024 | .checkentry = hashlimit_mt_check_v1, |
| 1025 | .destroy = hashlimit_mt_destroy_v1, |
| 1026 | .me = THIS_MODULE, |
| 1027 | }, |
| 1028 | { |
| 1029 | .name = "hashlimit" , |
| 1030 | .revision = 2, |
| 1031 | .family = NFPROTO_IPV6, |
| 1032 | .match = hashlimit_mt_v2, |
| 1033 | .matchsize = sizeof(struct xt_hashlimit_mtinfo2), |
| 1034 | .usersize = offsetof(struct xt_hashlimit_mtinfo2, hinfo), |
| 1035 | .checkentry = hashlimit_mt_check_v2, |
| 1036 | .destroy = hashlimit_mt_destroy_v2, |
| 1037 | .me = THIS_MODULE, |
| 1038 | }, |
| 1039 | { |
| 1040 | .name = "hashlimit" , |
| 1041 | .revision = 3, |
| 1042 | .family = NFPROTO_IPV6, |
| 1043 | .match = hashlimit_mt, |
| 1044 | .matchsize = sizeof(struct xt_hashlimit_mtinfo3), |
| 1045 | .usersize = offsetof(struct xt_hashlimit_mtinfo3, hinfo), |
| 1046 | .checkentry = hashlimit_mt_check, |
| 1047 | .destroy = hashlimit_mt_destroy, |
| 1048 | .me = THIS_MODULE, |
| 1049 | }, |
| 1050 | #endif |
| 1051 | }; |
| 1052 | |
| 1053 | /* PROC stuff */ |
| 1054 | static void *dl_seq_start(struct seq_file *s, loff_t *pos) |
| 1055 | __acquires(htable->lock) |
| 1056 | { |
| 1057 | struct xt_hashlimit_htable *htable = pde_data(inode: file_inode(f: s->file)); |
| 1058 | unsigned int *bucket; |
| 1059 | |
| 1060 | spin_lock_bh(lock: &htable->lock); |
| 1061 | if (*pos >= htable->cfg.size) |
| 1062 | return NULL; |
| 1063 | |
| 1064 | bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC); |
| 1065 | if (!bucket) |
| 1066 | return ERR_PTR(error: -ENOMEM); |
| 1067 | |
| 1068 | *bucket = *pos; |
| 1069 | return bucket; |
| 1070 | } |
| 1071 | |
| 1072 | static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos) |
| 1073 | { |
| 1074 | struct xt_hashlimit_htable *htable = pde_data(inode: file_inode(f: s->file)); |
| 1075 | unsigned int *bucket = v; |
| 1076 | |
| 1077 | *pos = ++(*bucket); |
| 1078 | if (*pos >= htable->cfg.size) { |
| 1079 | kfree(objp: v); |
| 1080 | return NULL; |
| 1081 | } |
| 1082 | return bucket; |
| 1083 | } |
| 1084 | |
| 1085 | static void dl_seq_stop(struct seq_file *s, void *v) |
| 1086 | __releases(htable->lock) |
| 1087 | { |
| 1088 | struct xt_hashlimit_htable *htable = pde_data(inode: file_inode(f: s->file)); |
| 1089 | unsigned int *bucket = v; |
| 1090 | |
| 1091 | if (!IS_ERR(ptr: bucket)) |
| 1092 | kfree(objp: bucket); |
| 1093 | spin_unlock_bh(lock: &htable->lock); |
| 1094 | } |
| 1095 | |
| 1096 | static void dl_seq_print(struct dsthash_ent *ent, u_int8_t family, |
| 1097 | struct seq_file *s) |
| 1098 | { |
| 1099 | switch (family) { |
| 1100 | case NFPROTO_IPV4: |
| 1101 | seq_printf(m: s, fmt: "%ld %pI4:%u->%pI4:%u %llu %llu %llu\n" , |
| 1102 | (long)(ent->expires - jiffies)/HZ, |
| 1103 | &ent->dst.ip.src, |
| 1104 | ntohs(ent->dst.src_port), |
| 1105 | &ent->dst.ip.dst, |
| 1106 | ntohs(ent->dst.dst_port), |
| 1107 | ent->rateinfo.credit, ent->rateinfo.credit_cap, |
| 1108 | ent->rateinfo.cost); |
| 1109 | break; |
| 1110 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) |
| 1111 | case NFPROTO_IPV6: |
| 1112 | seq_printf(m: s, fmt: "%ld %pI6:%u->%pI6:%u %llu %llu %llu\n" , |
| 1113 | (long)(ent->expires - jiffies)/HZ, |
| 1114 | &ent->dst.ip6.src, |
| 1115 | ntohs(ent->dst.src_port), |
| 1116 | &ent->dst.ip6.dst, |
| 1117 | ntohs(ent->dst.dst_port), |
| 1118 | ent->rateinfo.credit, ent->rateinfo.credit_cap, |
| 1119 | ent->rateinfo.cost); |
| 1120 | break; |
| 1121 | #endif |
| 1122 | default: |
| 1123 | BUG(); |
| 1124 | } |
| 1125 | } |
| 1126 | |
| 1127 | static int dl_seq_real_show_v2(struct dsthash_ent *ent, u_int8_t family, |
| 1128 | struct seq_file *s) |
| 1129 | { |
| 1130 | struct xt_hashlimit_htable *ht = pde_data(inode: file_inode(f: s->file)); |
| 1131 | |
| 1132 | spin_lock(lock: &ent->lock); |
| 1133 | /* recalculate to show accurate numbers */ |
| 1134 | rateinfo_recalc(dh: ent, now: jiffies, mode: ht->cfg.mode, revision: 2); |
| 1135 | |
| 1136 | dl_seq_print(ent, family, s); |
| 1137 | |
| 1138 | spin_unlock(lock: &ent->lock); |
| 1139 | return seq_has_overflowed(m: s); |
| 1140 | } |
| 1141 | |
| 1142 | static int dl_seq_real_show_v1(struct dsthash_ent *ent, u_int8_t family, |
| 1143 | struct seq_file *s) |
| 1144 | { |
| 1145 | struct xt_hashlimit_htable *ht = pde_data(inode: file_inode(f: s->file)); |
| 1146 | |
| 1147 | spin_lock(lock: &ent->lock); |
| 1148 | /* recalculate to show accurate numbers */ |
| 1149 | rateinfo_recalc(dh: ent, now: jiffies, mode: ht->cfg.mode, revision: 1); |
| 1150 | |
| 1151 | dl_seq_print(ent, family, s); |
| 1152 | |
| 1153 | spin_unlock(lock: &ent->lock); |
| 1154 | return seq_has_overflowed(m: s); |
| 1155 | } |
| 1156 | |
| 1157 | static int dl_seq_real_show(struct dsthash_ent *ent, u_int8_t family, |
| 1158 | struct seq_file *s) |
| 1159 | { |
| 1160 | struct xt_hashlimit_htable *ht = pde_data(inode: file_inode(f: s->file)); |
| 1161 | |
| 1162 | spin_lock(lock: &ent->lock); |
| 1163 | /* recalculate to show accurate numbers */ |
| 1164 | rateinfo_recalc(dh: ent, now: jiffies, mode: ht->cfg.mode, revision: 3); |
| 1165 | |
| 1166 | dl_seq_print(ent, family, s); |
| 1167 | |
| 1168 | spin_unlock(lock: &ent->lock); |
| 1169 | return seq_has_overflowed(m: s); |
| 1170 | } |
| 1171 | |
| 1172 | static int dl_seq_show_v2(struct seq_file *s, void *v) |
| 1173 | { |
| 1174 | struct xt_hashlimit_htable *htable = pde_data(inode: file_inode(f: s->file)); |
| 1175 | unsigned int *bucket = (unsigned int *)v; |
| 1176 | struct dsthash_ent *ent; |
| 1177 | |
| 1178 | if (!hlist_empty(h: &htable->hash[*bucket])) { |
| 1179 | hlist_for_each_entry(ent, &htable->hash[*bucket], node) |
| 1180 | if (dl_seq_real_show_v2(ent, family: htable->family, s)) |
| 1181 | return -1; |
| 1182 | } |
| 1183 | return 0; |
| 1184 | } |
| 1185 | |
| 1186 | static int dl_seq_show_v1(struct seq_file *s, void *v) |
| 1187 | { |
| 1188 | struct xt_hashlimit_htable *htable = pde_data(inode: file_inode(f: s->file)); |
| 1189 | unsigned int *bucket = v; |
| 1190 | struct dsthash_ent *ent; |
| 1191 | |
| 1192 | if (!hlist_empty(h: &htable->hash[*bucket])) { |
| 1193 | hlist_for_each_entry(ent, &htable->hash[*bucket], node) |
| 1194 | if (dl_seq_real_show_v1(ent, family: htable->family, s)) |
| 1195 | return -1; |
| 1196 | } |
| 1197 | return 0; |
| 1198 | } |
| 1199 | |
| 1200 | static int dl_seq_show(struct seq_file *s, void *v) |
| 1201 | { |
| 1202 | struct xt_hashlimit_htable *htable = pde_data(inode: file_inode(f: s->file)); |
| 1203 | unsigned int *bucket = v; |
| 1204 | struct dsthash_ent *ent; |
| 1205 | |
| 1206 | if (!hlist_empty(h: &htable->hash[*bucket])) { |
| 1207 | hlist_for_each_entry(ent, &htable->hash[*bucket], node) |
| 1208 | if (dl_seq_real_show(ent, family: htable->family, s)) |
| 1209 | return -1; |
| 1210 | } |
| 1211 | return 0; |
| 1212 | } |
| 1213 | |
| 1214 | static const struct seq_operations dl_seq_ops_v1 = { |
| 1215 | .start = dl_seq_start, |
| 1216 | .next = dl_seq_next, |
| 1217 | .stop = dl_seq_stop, |
| 1218 | .show = dl_seq_show_v1 |
| 1219 | }; |
| 1220 | |
| 1221 | static const struct seq_operations dl_seq_ops_v2 = { |
| 1222 | .start = dl_seq_start, |
| 1223 | .next = dl_seq_next, |
| 1224 | .stop = dl_seq_stop, |
| 1225 | .show = dl_seq_show_v2 |
| 1226 | }; |
| 1227 | |
| 1228 | static const struct seq_operations dl_seq_ops = { |
| 1229 | .start = dl_seq_start, |
| 1230 | .next = dl_seq_next, |
| 1231 | .stop = dl_seq_stop, |
| 1232 | .show = dl_seq_show |
| 1233 | }; |
| 1234 | |
| 1235 | static int __net_init hashlimit_proc_net_init(struct net *net) |
| 1236 | { |
| 1237 | struct hashlimit_net *hashlimit_net = hashlimit_pernet(net); |
| 1238 | |
| 1239 | hashlimit_net->ipt_hashlimit = proc_mkdir("ipt_hashlimit" , net->proc_net); |
| 1240 | if (!hashlimit_net->ipt_hashlimit) |
| 1241 | return -ENOMEM; |
| 1242 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) |
| 1243 | hashlimit_net->ip6t_hashlimit = proc_mkdir("ip6t_hashlimit" , net->proc_net); |
| 1244 | if (!hashlimit_net->ip6t_hashlimit) { |
| 1245 | remove_proc_entry("ipt_hashlimit" , net->proc_net); |
| 1246 | return -ENOMEM; |
| 1247 | } |
| 1248 | #endif |
| 1249 | return 0; |
| 1250 | } |
| 1251 | |
| 1252 | static void __net_exit hashlimit_proc_net_exit(struct net *net) |
| 1253 | { |
| 1254 | struct xt_hashlimit_htable *hinfo; |
| 1255 | struct hashlimit_net *hashlimit_net = hashlimit_pernet(net); |
| 1256 | |
| 1257 | /* hashlimit_net_exit() is called before hashlimit_mt_destroy(). |
| 1258 | * Make sure that the parent ipt_hashlimit and ip6t_hashlimit proc |
| 1259 | * entries is empty before trying to remove it. |
| 1260 | */ |
| 1261 | mutex_lock(&hashlimit_mutex); |
| 1262 | hlist_for_each_entry(hinfo, &hashlimit_net->htables, node) |
| 1263 | htable_remove_proc_entry(hinfo); |
| 1264 | hashlimit_net->ipt_hashlimit = NULL; |
| 1265 | hashlimit_net->ip6t_hashlimit = NULL; |
| 1266 | mutex_unlock(lock: &hashlimit_mutex); |
| 1267 | |
| 1268 | remove_proc_entry("ipt_hashlimit" , net->proc_net); |
| 1269 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) |
| 1270 | remove_proc_entry("ip6t_hashlimit" , net->proc_net); |
| 1271 | #endif |
| 1272 | } |
| 1273 | |
| 1274 | static int __net_init hashlimit_net_init(struct net *net) |
| 1275 | { |
| 1276 | struct hashlimit_net *hashlimit_net = hashlimit_pernet(net); |
| 1277 | |
| 1278 | INIT_HLIST_HEAD(&hashlimit_net->htables); |
| 1279 | return hashlimit_proc_net_init(net); |
| 1280 | } |
| 1281 | |
| 1282 | static void __net_exit hashlimit_net_exit(struct net *net) |
| 1283 | { |
| 1284 | hashlimit_proc_net_exit(net); |
| 1285 | } |
| 1286 | |
| 1287 | static struct pernet_operations hashlimit_net_ops = { |
| 1288 | .init = hashlimit_net_init, |
| 1289 | .exit = hashlimit_net_exit, |
| 1290 | .id = &hashlimit_net_id, |
| 1291 | .size = sizeof(struct hashlimit_net), |
| 1292 | }; |
| 1293 | |
| 1294 | static int __init hashlimit_mt_init(void) |
| 1295 | { |
| 1296 | int err; |
| 1297 | |
| 1298 | err = register_pernet_subsys(&hashlimit_net_ops); |
| 1299 | if (err < 0) |
| 1300 | return err; |
| 1301 | err = xt_register_matches(match: hashlimit_mt_reg, |
| 1302 | ARRAY_SIZE(hashlimit_mt_reg)); |
| 1303 | if (err < 0) |
| 1304 | goto err1; |
| 1305 | |
| 1306 | err = -ENOMEM; |
| 1307 | hashlimit_cachep = kmem_cache_create("xt_hashlimit" , |
| 1308 | sizeof(struct dsthash_ent), 0, 0, |
| 1309 | NULL); |
| 1310 | if (!hashlimit_cachep) { |
| 1311 | pr_warn("unable to create slab cache\n" ); |
| 1312 | goto err2; |
| 1313 | } |
| 1314 | return 0; |
| 1315 | |
| 1316 | err2: |
| 1317 | xt_unregister_matches(match: hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg)); |
| 1318 | err1: |
| 1319 | unregister_pernet_subsys(&hashlimit_net_ops); |
| 1320 | return err; |
| 1321 | |
| 1322 | } |
| 1323 | |
| 1324 | static void __exit hashlimit_mt_exit(void) |
| 1325 | { |
| 1326 | xt_unregister_matches(match: hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg)); |
| 1327 | unregister_pernet_subsys(&hashlimit_net_ops); |
| 1328 | |
| 1329 | rcu_barrier(); |
| 1330 | kmem_cache_destroy(s: hashlimit_cachep); |
| 1331 | } |
| 1332 | |
| 1333 | module_init(hashlimit_mt_init); |
| 1334 | module_exit(hashlimit_mt_exit); |
| 1335 | |