| 1 | |
| 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 <[email protected]>
|
| 7 | * (C) 2006-2012 Patrick McHardy <[email protected]>
|
| 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 <[email protected]>");
|
| 48 | MODULE_AUTHOR("Jan Engelhardt <[email protected]>");
|
| 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, hashlimit_net_id);
|
| 63 | }
|
| 64 |
|
| 65 |
|
| 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 |
|
| 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 |
|
| 90 | struct hlist_node node;
|
| 91 | struct dsthash_dst dst;
|
| 92 |
|
| 93 |
|
| 94 | spinlock_t lock;
|
| 95 | unsigned long expires;
|
| 96 | struct {
|
| 97 | unsigned long prev;
|
| 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;
|
| 117 | refcount_t use;
|
| 118 | u_int8_t family;
|
| 119 | bool rnd_initialized;
|
| 120 | bool ratematch;
|
| 121 |
|
| 122 | struct hashlimit_cfg3 cfg;
|
| 123 |
|
| 124 |
|
| 125 | spinlock_t lock;
|
| 126 | u_int32_t rnd;
|
| 127 | unsigned int count;
|
| 128 | struct delayed_work gc_work;
|
| 129 |
|
| 130 |
|
| 131 | struct proc_dir_entry *pde;
|
| 132 | const char *name;
|
| 133 | struct net *net;
|
| 134 |
|
| 135 | struct hlist_head hash[];
|
| 136 | };
|
| 137 |
|
| 138 | static int
|
| 139 | cfg_copy(struct hashlimit_cfg3 *to, const void *from, int revision)
|
| 140 | {
|
| 141 | if (revision == 1) {
|
| 142 | struct hashlimit_cfg1 *cfg = (struct hashlimit_cfg1 *)from;
|
| 143 |
|
| 144 | to->mode = cfg->mode;
|
| 145 | to->avg = cfg->avg;
|
| 146 | to->burst = cfg->burst;
|
| 147 | to->size = cfg->size;
|
| 148 | to->max = cfg->max;
|
| 149 | to->gc_interval = cfg->gc_interval;
|
| 150 | to->expire = cfg->expire;
|
| 151 | to->srcmask = cfg->srcmask;
|
| 152 | to->dstmask = cfg->dstmask;
|
| 153 | } else if (revision == 2) {
|
| 154 | struct hashlimit_cfg2 *cfg = (struct hashlimit_cfg2 *)from;
|
| 155 |
|
| 156 | to->mode = cfg->mode;
|
| 157 | to->avg = cfg->avg;
|
| 158 | to->burst = cfg->burst;
|
| 159 | to->size = cfg->size;
|
| 160 | to->max = cfg->max;
|
| 161 | to->gc_interval = cfg->gc_interval;
|
| 162 | to->expire = cfg->expire;
|
| 163 | to->srcmask = cfg->srcmask;
|
| 164 | to->dstmask = cfg->dstmask;
|
| 165 | } else if (revision == 3) {
|
| 166 | memcpy(to, from, sizeof(struct hashlimit_cfg3));
|
| 167 | } else {
|
| 168 | return -EINVAL;
|
| 169 | }
|
| 170 |
|
| 171 | return 0;
|
| 172 | }
|
| 173 |
|
| 174 | static DEFINE_MUTEX(hashlimit_mutex);
|
| 175 | static struct kmem_cache *hashlimit_cachep __read_mostly;
|
| 176 |
|
| 177 | static inline bool dst_cmp(const struct dsthash_ent *ent,
|
| 178 | const struct dsthash_dst *b)
|
| 179 | {
|
| 180 | return !memcmp(&ent->dst, b, sizeof(ent->dst));
|
| 181 | }
|
| 182 |
|
| 183 | static u_int32_t
|
| 184 | hash_dst(const struct xt_hashlimit_htable *ht, const struct dsthash_dst *dst)
|
| 185 | {
|
| 186 | u_int32_t hash = jhash2((const u32 *)dst,
|
| 187 | sizeof(*dst)/sizeof(u32),
|
| 188 | ht->rnd);
|
| 189 | |
| 190 | * Instead of returning hash % ht->cfg.size (implying a divide)
|
| 191 | * we return the high 32 bits of the (hash * ht->cfg.size) that will
|
| 192 | * give results between [0 and cfg.size-1] and same hash distribution,
|
| 193 | * but using a multiply, less expensive than a divide
|
| 194 |
|
| 195 | return reciprocal_scale(hash, ht->cfg.size);
|
| 196 | }
|
| 197 |
|
| 198 | static struct dsthash_ent *
|
| 199 | dsthash_find(const struct xt_hashlimit_htable *ht,
|
| 200 | const struct dsthash_dst *dst)
|
| 201 | {
|
| 202 | struct dsthash_ent *ent;
|
| 203 | u_int32_t hash = hash_dst(ht, dst);
|
| 204 |
|
| 205 | if (!hlist_empty(&ht->hash[hash])) {
|
| 206 | hlist_for_each_entry_rcu(ent, &ht->hash[hash], node)
|
| 207 | if (dst_cmp(ent, dst)) {
|
| 208 | spin_lock(&ent->lock);
|
| 209 | return ent;
|
| 210 | }
|
| 211 | }
|
| 212 | return NULL;
|
| 213 | }
|
| 214 |
|
| 215 |
|
| 216 | static struct dsthash_ent *
|
| 217 | dsthash_alloc_init(struct xt_hashlimit_htable *ht,
|
| 218 | const struct dsthash_dst *dst, bool *race)
|
| 219 | {
|
| 220 | struct dsthash_ent *ent;
|
| 221 |
|
| 222 | spin_lock(&ht->lock);
|
| 223 |
|
| 224 | |
| 225 | * hashtable, double check if this packet lost race.
|
| 226 |
|
| 227 | ent = dsthash_find(ht, dst);
|
| 228 | if (ent != NULL) {
|
| 229 | spin_unlock(&ht->lock);
|
| 230 | *race = true;
|
| 231 | return ent;
|
| 232 | }
|
| 233 |
|
| 234 | |
| 235 | * the first hashtable entry
|
| 236 | if (unlikely(!ht->rnd_initialized)) {
|
| 237 | get_random_bytes(&ht->rnd, sizeof(ht->rnd));
|
| 238 | ht->rnd_initialized = true;
|
| 239 | }
|
| 240 |
|
| 241 | if (ht->cfg.max && ht->count >= ht->cfg.max) {
|
| 242 |
|
| 243 | net_err_ratelimited("max count of %u reached\n", ht->cfg.max);
|
| 244 | ent = NULL;
|
| 245 | } else
|
| 246 | ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC);
|
| 247 | if (ent) {
|
| 248 | memcpy(&ent->dst, dst, sizeof(ent->dst));
|
| 249 | spin_lock_init(&ent->lock);
|
| 250 |
|
| 251 | spin_lock(&ent->lock);
|
| 252 | hlist_add_head_rcu(&ent->node, &ht->hash[hash_dst(ht, dst)]);
|
| 253 | ht->count++;
|
| 254 | }
|
| 255 | spin_unlock(&ht->lock);
|
| 256 | return ent;
|
| 257 | }
|
| 258 |
|
| 259 | static void dsthash_free_rcu(struct rcu_head *head)
|
| 260 | {
|
| 261 | struct dsthash_ent *ent = container_of(head, struct dsthash_ent, rcu);
|
| 262 |
|
| 263 | kmem_cache_free(hashlimit_cachep, ent);
|
| 264 | }
|
| 265 |
|
| 266 | static inline void
|
| 267 | dsthash_free(struct xt_hashlimit_htable *ht, struct dsthash_ent *ent)
|
| 268 | {
|
| 269 | hlist_del_rcu(&ent->node);
|
| 270 | call_rcu(&ent->rcu, dsthash_free_rcu);
|
| 271 | ht->count--;
|
| 272 | }
|
| 273 | static void htable_gc(struct work_struct *work);
|
| 274 |
|
| 275 | static int htable_create(struct net *net, struct hashlimit_cfg3 *cfg,
|
| 276 | const char *name, u_int8_t family,
|
| 277 | struct xt_hashlimit_htable **out_hinfo,
|
| 278 | int revision)
|
| 279 | {
|
| 280 | struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
|
| 281 | struct xt_hashlimit_htable *hinfo;
|
| 282 | const struct seq_operations *ops;
|
| 283 | unsigned int size, i;
|
| 284 | unsigned long nr_pages = totalram_pages();
|
| 285 | int ret;
|
| 286 |
|
| 287 | if (cfg->size) {
|
| 288 | size = cfg->size;
|
| 289 | } else {
|
| 290 | size = (nr_pages << PAGE_SHIFT) / 16384 /
|
| 291 | sizeof(struct hlist_head);
|
| 292 | if (nr_pages > 1024 * 1024 * 1024 / PAGE_SIZE)
|
| 293 | size = 8192;
|
| 294 | if (size < 16)
|
| 295 | size = 16;
|
| 296 | }
|
| 297 | hinfo = kvmalloc_flex(*hinfo, hash, size);
|
| 298 | if (hinfo == NULL)
|
| 299 | return -ENOMEM;
|
| 300 | *out_hinfo = hinfo;
|
| 301 |
|
| 302 |
|
| 303 | ret = cfg_copy(&hinfo->cfg, (void *)cfg, 3);
|
| 304 | if (ret) {
|
| 305 | kvfree(hinfo);
|
| 306 | return ret;
|
| 307 | }
|
| 308 |
|
| 309 | hinfo->cfg.size = size;
|
| 310 | if (hinfo->cfg.max == 0)
|
| 311 | hinfo->cfg.max = 8 * hinfo->cfg.size;
|
| 312 | else if (hinfo->cfg.max < hinfo->cfg.size)
|
| 313 | hinfo->cfg.max = hinfo->cfg.size;
|
| 314 |
|
| 315 | for (i = 0; i < hinfo->cfg.size; i++)
|
| 316 | INIT_HLIST_HEAD(&hinfo->hash[i]);
|
| 317 |
|
| 318 | refcount_set(&hinfo->use, 1);
|
| 319 | hinfo->count = 0;
|
| 320 | hinfo->family = family;
|
| 321 | hinfo->rnd_initialized = false;
|
| 322 | hinfo->name = kstrdup(name, GFP_KERNEL);
|
| 323 | if (!hinfo->name) {
|
| 324 | kvfree(hinfo);
|
| 325 | return -ENOMEM;
|
| 326 | }
|
| 327 | hinfo->ratematch = !!(cfg->mode & XT_HASHLIMIT_RATE_MATCH);
|
| 328 | spin_lock_init(&hinfo->lock);
|
| 329 |
|
| 330 | switch (revision) {
|
| 331 | case 1:
|
| 332 | ops = &dl_seq_ops_v1;
|
| 333 | break;
|
| 334 | case 2:
|
| 335 | ops = &dl_seq_ops_v2;
|
| 336 | break;
|
| 337 | default:
|
| 338 | ops = &dl_seq_ops;
|
| 339 | }
|
| 340 |
|
| 341 | hinfo->pde = proc_create_seq_data(name, 0,
|
| 342 | (family == NFPROTO_IPV4) ?
|
| 343 | hashlimit_net->ipt_hashlimit : hashlimit_net->ip6t_hashlimit,
|
| 344 | ops, hinfo);
|
| 345 | if (hinfo->pde == NULL) {
|
| 346 | kfree(hinfo->name);
|
| 347 | kvfree(hinfo);
|
| 348 | return -ENOMEM;
|
| 349 | }
|
| 350 | hinfo->net = net;
|
| 351 |
|
| 352 | INIT_DEFERRABLE_WORK(&hinfo->gc_work, htable_gc);
|
| 353 | queue_delayed_work(system_power_efficient_wq, &hinfo->gc_work,
|
| 354 | msecs_to_jiffies(hinfo->cfg.gc_interval));
|
| 355 |
|
| 356 | hlist_add_head(&hinfo->node, &hashlimit_net->htables);
|
| 357 |
|
| 358 | return 0;
|
| 359 | }
|
| 360 |
|
| 361 | static void htable_selective_cleanup(struct xt_hashlimit_htable *ht, bool select_all)
|
| 362 | {
|
| 363 | unsigned int i;
|
| 364 |
|
| 365 | for (i = 0; i < ht->cfg.size; i++) {
|
| 366 | struct hlist_head *head = &ht->hash[i];
|
| 367 | struct dsthash_ent *dh;
|
| 368 | struct hlist_node *n;
|
| 369 |
|
| 370 | if (hlist_empty(head))
|
| 371 | continue;
|
| 372 |
|
| 373 | spin_lock_bh(&ht->lock);
|
| 374 | hlist_for_each_entry_safe(dh, n, head, node) {
|
| 375 | if (time_after_eq(jiffies, dh->expires) || select_all)
|
| 376 | dsthash_free(ht, dh);
|
| 377 | }
|
| 378 | spin_unlock_bh(&ht->lock);
|
| 379 | cond_resched();
|
| 380 | }
|
| 381 | }
|
| 382 |
|
| 383 | static void htable_gc(struct work_struct *work)
|
| 384 | {
|
| 385 | struct xt_hashlimit_htable *ht;
|
| 386 |
|
| 387 | ht = container_of(work, struct xt_hashlimit_htable, gc_work.work);
|
| 388 |
|
| 389 | htable_selective_cleanup(ht, false);
|
| 390 |
|
| 391 | queue_delayed_work(system_power_efficient_wq,
|
| 392 | &ht->gc_work, msecs_to_jiffies(ht->cfg.gc_interval));
|
| 393 | }
|
| 394 |
|
| 395 | static void htable_remove_proc_entry(struct xt_hashlimit_htable *hinfo)
|
| 396 | {
|
| 397 | struct hashlimit_net *hashlimit_net = hashlimit_pernet(hinfo->net);
|
| 398 | struct proc_dir_entry *parent;
|
| 399 |
|
| 400 | if (hinfo->family == NFPROTO_IPV4)
|
| 401 | parent = hashlimit_net->ipt_hashlimit;
|
| 402 | else
|
| 403 | parent = hashlimit_net->ip6t_hashlimit;
|
| 404 |
|
| 405 | if (parent != NULL)
|
| 406 | remove_proc_entry(hinfo->name, parent);
|
| 407 | }
|
| 408 |
|
| 409 | static struct xt_hashlimit_htable *htable_find_get(struct net *net,
|
| 410 | const char *name,
|
| 411 | u_int8_t family)
|
| 412 | {
|
| 413 | struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
|
| 414 | struct xt_hashlimit_htable *hinfo;
|
| 415 |
|
| 416 | hlist_for_each_entry(hinfo, &hashlimit_net->htables, node) {
|
| 417 | if (!strcmp(name, hinfo->name) &&
|
| 418 | hinfo->family == family) {
|
| 419 | refcount_inc(&hinfo->use);
|
| 420 | return hinfo;
|
| 421 | }
|
| 422 | }
|
| 423 | return NULL;
|
| 424 | }
|
| 425 |
|
| 426 | static void htable_put(struct xt_hashlimit_htable *hinfo)
|
| 427 | {
|
| 428 | if (refcount_dec_and_mutex_lock(&hinfo->use, &hashlimit_mutex)) {
|
| 429 | hlist_del(&hinfo->node);
|
| 430 | htable_remove_proc_entry(hinfo);
|
| 431 | mutex_unlock(&hashlimit_mutex);
|
| 432 |
|
| 433 | cancel_delayed_work_sync(&hinfo->gc_work);
|
| 434 | htable_selective_cleanup(hinfo, true);
|
| 435 | kfree(hinfo->name);
|
| 436 | kvfree(hinfo);
|
| 437 | }
|
| 438 | }
|
| 439 |
|
| 440 | |
| 441 | * see net/sched/sch_tbf.c in the linux source tree
|
| 442 |
|
| 443 |
|
| 444 | |
| 445 | this algorithm. The `average rate' in jiffies becomes your initial
|
| 446 | amount of credit `credit' and the most credit you can ever have
|
| 447 | `credit_cap'. The `peak rate' becomes the cost of passing the
|
| 448 | test, `cost'.
|
| 449 |
|
| 450 | `prev' tracks the last packet hit: you gain one credit per jiffy.
|
| 451 | If you get credit balance more than this, the extra credit is
|
| 452 | discarded. Every time the match passes, you lose `cost' credits;
|
| 453 | if you don't have that many, the test fails.
|
| 454 |
|
| 455 | See Alexey's formal explanation in net/sched/sch_tbf.c.
|
| 456 |
|
| 457 | To get the maximum range, we multiply by this factor (ie. you get N
|
| 458 | credits per jiffy). We want to allow a rate as low as 1 per day
|
| 459 | (slowest userspace tool allows), which means
|
| 460 | CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
|
| 461 |
|
| 462 | #define MAX_CPJ_v1 (0xFFFFFFFF / (HZ*60*60*24))
|
| 463 | #define MAX_CPJ (0xFFFFFFFFFFFFFFFFULL / (HZ*60*60*24))
|
| 464 |
|
| 465 | |
| 466 | * us the power of 2 below the theoretical max, so GCC simply does a
|
| 467 | * shift.
|
| 468 | #define _POW2_BELOW2(x) ((x)|((x)>>1))
|
| 469 | #define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
|
| 470 | #define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
|
| 471 | #define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
|
| 472 | #define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
|
| 473 | #define _POW2_BELOW64(x) (_POW2_BELOW32(x)|_POW2_BELOW32((x)>>32))
|
| 474 | #define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
|
| 475 | #define POW2_BELOW64(x) ((_POW2_BELOW64(x)>>1) + 1)
|
| 476 |
|
| 477 | #define CREDITS_PER_JIFFY POW2_BELOW64(MAX_CPJ)
|
| 478 | #define CREDITS_PER_JIFFY_v1 POW2_BELOW32(MAX_CPJ_v1)
|
| 479 |
|
| 480 | |
| 481 | * credit_cap is used as a counter that tells us how many times we can
|
| 482 | * refill the "credits available" counter when it becomes empty.
|
| 483 |
|
| 484 | #define MAX_CPJ_BYTES (0xFFFFFFFF / HZ)
|
| 485 | #define CREDITS_PER_JIFFY_BYTES POW2_BELOW32(MAX_CPJ_BYTES)
|
| 486 |
|
| 487 | static u32 xt_hashlimit_len_to_chunks(u32 len)
|
| 488 | {
|
| 489 | return (len >> XT_HASHLIMIT_BYTE_SHIFT) + 1;
|
| 490 | }
|
| 491 |
|
| 492 |
|
| 493 | static u64 user2credits(u64 user, int revision)
|
| 494 | {
|
| 495 | u64 scale = (revision == 1) ?
|
| 496 | XT_HASHLIMIT_SCALE : XT_HASHLIMIT_SCALE_v2;
|
| 497 | u64 cpj = (revision == 1) ?
|
| 498 | CREDITS_PER_JIFFY_v1 : CREDITS_PER_JIFFY;
|
| 499 |
|
| 500 |
|
| 501 | if (scale >= HZ * cpj)
|
| 502 | return div64_u64(user, div64_u64(scale, HZ * cpj));
|
| 503 |
|
| 504 | return user * div64_u64(HZ * cpj, scale);
|
| 505 | }
|
| 506 |
|
| 507 | static u32 user2credits_byte(u32 user)
|
| 508 | {
|
| 509 | u64 us = user;
|
| 510 | us *= HZ * CREDITS_PER_JIFFY_BYTES;
|
| 511 | return (u32) (us >> 32);
|
| 512 | }
|
| 513 |
|
| 514 | static u64 user2rate(u64 user)
|
| 515 | {
|
| 516 | if (user != 0) {
|
| 517 | return div64_u64(XT_HASHLIMIT_SCALE_v2, user);
|
| 518 | } else {
|
| 519 | pr_info_ratelimited("invalid rate from userspace: %llu\n",
|
| 520 | user);
|
| 521 | return 0;
|
| 522 | }
|
| 523 | }
|
| 524 |
|
| 525 | static u64 user2rate_bytes(u32 user)
|
| 526 | {
|
| 527 | u64 r;
|
| 528 |
|
| 529 | r = user ? U32_MAX / user : U32_MAX;
|
| 530 | return (r - 1) << XT_HASHLIMIT_BYTE_SHIFT;
|
| 531 | }
|
| 532 |
|
| 533 | static void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now,
|
| 534 | u32 mode, int revision)
|
| 535 | {
|
| 536 | unsigned long delta = now - dh->rateinfo.prev;
|
| 537 | u64 cap, cpj;
|
| 538 |
|
| 539 | if (delta == 0)
|
| 540 | return;
|
| 541 |
|
| 542 | if (revision >= 3 && mode & XT_HASHLIMIT_RATE_MATCH) {
|
| 543 | u64 interval = dh->rateinfo.interval * HZ;
|
| 544 |
|
| 545 | if (delta < interval)
|
| 546 | return;
|
| 547 |
|
| 548 | dh->rateinfo.prev = now;
|
| 549 | dh->rateinfo.prev_window =
|
| 550 | ((dh->rateinfo.current_rate * interval) >
|
| 551 | (delta * dh->rateinfo.rate));
|
| 552 | dh->rateinfo.current_rate = 0;
|
| 553 |
|
| 554 | return;
|
| 555 | }
|
| 556 |
|
| 557 | dh->rateinfo.prev = now;
|
| 558 |
|
| 559 | if (mode & XT_HASHLIMIT_BYTES) {
|
| 560 | u64 tmp = dh->rateinfo.credit;
|
| 561 | dh->rateinfo.credit += CREDITS_PER_JIFFY_BYTES * delta;
|
| 562 | cap = CREDITS_PER_JIFFY_BYTES * HZ;
|
| 563 | if (tmp >= dh->rateinfo.credit) {
|
| 564 | dh->rateinfo.credit = cap;
|
| 565 | return;
|
| 566 | }
|
| 567 | } else {
|
| 568 | cpj = (revision == 1) ?
|
| 569 | CREDITS_PER_JIFFY_v1 : CREDITS_PER_JIFFY;
|
| 570 | dh->rateinfo.credit += delta * cpj;
|
| 571 | cap = dh->rateinfo.credit_cap;
|
| 572 | }
|
| 573 | if (dh->rateinfo.credit > cap)
|
| 574 | dh->rateinfo.credit = cap;
|
| 575 | }
|
| 576 |
|
| 577 | static void rateinfo_init(struct dsthash_ent *dh,
|
| 578 | struct xt_hashlimit_htable *hinfo, int revision)
|
| 579 | {
|
| 580 | dh->rateinfo.prev = jiffies;
|
| 581 | if (revision >= 3 && hinfo->cfg.mode & XT_HASHLIMIT_RATE_MATCH) {
|
| 582 | dh->rateinfo.prev_window = 0;
|
| 583 | dh->rateinfo.current_rate = 0;
|
| 584 | if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) {
|
| 585 | dh->rateinfo.rate =
|
| 586 | user2rate_bytes((u32)hinfo->cfg.avg);
|
| 587 | if (hinfo->cfg.burst)
|
| 588 | dh->rateinfo.burst =
|
| 589 | hinfo->cfg.burst * dh->rateinfo.rate;
|
| 590 | else
|
| 591 | dh->rateinfo.burst = dh->rateinfo.rate;
|
| 592 | } else {
|
| 593 | dh->rateinfo.rate = user2rate(hinfo->cfg.avg);
|
| 594 | dh->rateinfo.burst =
|
| 595 | hinfo->cfg.burst + dh->rateinfo.rate;
|
| 596 | }
|
| 597 | dh->rateinfo.interval = hinfo->cfg.interval;
|
| 598 | } else if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) {
|
| 599 | dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
|
| 600 | dh->rateinfo.cost = user2credits_byte(hinfo->cfg.avg);
|
| 601 | dh->rateinfo.credit_cap = hinfo->cfg.burst;
|
| 602 | } else {
|
| 603 | dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
|
| 604 | hinfo->cfg.burst, revision);
|
| 605 | dh->rateinfo.cost = user2credits(hinfo->cfg.avg, revision);
|
| 606 | dh->rateinfo.credit_cap = dh->rateinfo.credit;
|
| 607 | }
|
| 608 | }
|
| 609 |
|
| 610 | static inline __be32 maskl(__be32 a, unsigned int l)
|
| 611 | {
|
| 612 | return l ? htonl(ntohl(a) & ~0 << (32 - l)) : 0;
|
| 613 | }
|
| 614 |
|
| 615 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
|
| 616 | static void hashlimit_ipv6_mask(__be32 *i, unsigned int p)
|
| 617 | {
|
| 618 | switch (p) {
|
| 619 | case 0 ... 31:
|
| 620 | i[0] = maskl(i[0], p);
|
| 621 | i[1] = i[2] = i[3] = 0;
|
| 622 | break;
|
| 623 | case 32 ... 63:
|
| 624 | i[1] = maskl(i[1], p - 32);
|
| 625 | i[2] = i[3] = 0;
|
| 626 | break;
|
| 627 | case 64 ... 95:
|
| 628 | i[2] = maskl(i[2], p - 64);
|
| 629 | i[3] = 0;
|
| 630 | break;
|
| 631 | case 96 ... 127:
|
| 632 | i[3] = maskl(i[3], p - 96);
|
| 633 | break;
|
| 634 | case 128:
|
| 635 | break;
|
| 636 | }
|
| 637 | }
|
| 638 | #endif
|
| 639 |
|
| 640 | static int
|
| 641 | hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo,
|
| 642 | struct dsthash_dst *dst,
|
| 643 | const struct sk_buff *skb, unsigned int protoff)
|
| 644 | {
|
| 645 | __be16 _ports[2], *ports;
|
| 646 | u8 nexthdr;
|
| 647 | int poff;
|
| 648 |
|
| 649 | memset(dst, 0, sizeof(*dst));
|
| 650 |
|
| 651 | switch (hinfo->family) {
|
| 652 | case NFPROTO_IPV4:
|
| 653 | if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
|
| 654 | dst->ip.dst = maskl(ip_hdr(skb)->daddr,
|
| 655 | hinfo->cfg.dstmask);
|
| 656 | if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
|
| 657 | dst->ip.src = maskl(ip_hdr(skb)->saddr,
|
| 658 | hinfo->cfg.srcmask);
|
| 659 |
|
| 660 | if (!(hinfo->cfg.mode &
|
| 661 | (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
|
| 662 | return 0;
|
| 663 | if (ntohs(ip_hdr(skb)->frag_off) & IP_OFFSET)
|
| 664 | return -1;
|
| 665 | nexthdr = ip_hdr(skb)->protocol;
|
| 666 | break;
|
| 667 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
|
| 668 | case NFPROTO_IPV6:
|
| 669 | {
|
| 670 | __be16 frag_off;
|
| 671 |
|
| 672 | if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) {
|
| 673 | memcpy(&dst->ip6.dst, &ipv6_hdr(skb)->daddr,
|
| 674 | sizeof(dst->ip6.dst));
|
| 675 | hashlimit_ipv6_mask(dst->ip6.dst, hinfo->cfg.dstmask);
|
| 676 | }
|
| 677 | if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) {
|
| 678 | memcpy(&dst->ip6.src, &ipv6_hdr(skb)->saddr,
|
| 679 | sizeof(dst->ip6.src));
|
| 680 | hashlimit_ipv6_mask(dst->ip6.src, hinfo->cfg.srcmask);
|
| 681 | }
|
| 682 |
|
| 683 | if (!(hinfo->cfg.mode &
|
| 684 | (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
|
| 685 | return 0;
|
| 686 | nexthdr = ipv6_hdr(skb)->nexthdr;
|
| 687 | protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr, &frag_off);
|
| 688 | if ((int)protoff < 0 || ntohs(frag_off) & IP6_OFFSET)
|
| 689 | return -1;
|
| 690 | break;
|
| 691 | }
|
| 692 | #endif
|
| 693 | default:
|
| 694 | BUG();
|
| 695 | return 0;
|
| 696 | }
|
| 697 |
|
| 698 | poff = proto_ports_offset(nexthdr);
|
| 699 | if (poff >= 0) {
|
| 700 | ports = skb_header_pointer(skb, protoff + poff, sizeof(_ports),
|
| 701 | &_ports);
|
| 702 | } else {
|
| 703 | _ports[0] = _ports[1] = 0;
|
| 704 | ports = _ports;
|
| 705 | }
|
| 706 | if (!ports)
|
| 707 | return -1;
|
| 708 | if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SPT)
|
| 709 | dst->src_port = ports[0];
|
| 710 | if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DPT)
|
| 711 | dst->dst_port = ports[1];
|
| 712 | return 0;
|
| 713 | }
|
| 714 |
|
| 715 | static u32 hashlimit_byte_cost(unsigned int len, struct dsthash_ent *dh)
|
| 716 | {
|
| 717 | u64 tmp = xt_hashlimit_len_to_chunks(len);
|
| 718 | tmp = tmp * dh->rateinfo.cost;
|
| 719 |
|
| 720 | if (unlikely(tmp > CREDITS_PER_JIFFY_BYTES * HZ))
|
| 721 | tmp = CREDITS_PER_JIFFY_BYTES * HZ;
|
| 722 |
|
| 723 | if (dh->rateinfo.credit < tmp && dh->rateinfo.credit_cap) {
|
| 724 | dh->rateinfo.credit_cap--;
|
| 725 | dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
|
| 726 | }
|
| 727 | return (u32) tmp;
|
| 728 | }
|
| 729 |
|
| 730 | static bool
|
| 731 | hashlimit_mt_common(const struct sk_buff *skb, struct xt_action_param *par,
|
| 732 | struct xt_hashlimit_htable *hinfo,
|
| 733 | const struct hashlimit_cfg3 *cfg, int revision)
|
| 734 | {
|
| 735 | unsigned long now = jiffies;
|
| 736 | struct dsthash_ent *dh;
|
| 737 | struct dsthash_dst dst;
|
| 738 | bool race = false;
|
| 739 | u64 cost;
|
| 740 |
|
| 741 | if (hashlimit_init_dst(hinfo, &dst, skb, par->thoff) < 0)
|
| 742 | goto hotdrop;
|
| 743 |
|
| 744 | local_bh_disable();
|
| 745 | dh = dsthash_find(hinfo, &dst);
|
| 746 | if (dh == NULL) {
|
| 747 | dh = dsthash_alloc_init(hinfo, &dst, &race);
|
| 748 | if (dh == NULL) {
|
| 749 | local_bh_enable();
|
| 750 | goto hotdrop;
|
| 751 | } else if (race) {
|
| 752 |
|
| 753 | dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
|
| 754 | rateinfo_recalc(dh, now, hinfo->cfg.mode, revision);
|
| 755 | } else {
|
| 756 | dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
|
| 757 | rateinfo_init(dh, hinfo, revision);
|
| 758 | }
|
| 759 | } else {
|
| 760 |
|
| 761 | dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
|
| 762 | rateinfo_recalc(dh, now, hinfo->cfg.mode, revision);
|
| 763 | }
|
| 764 |
|
| 765 | if (cfg->mode & XT_HASHLIMIT_RATE_MATCH) {
|
| 766 | cost = (cfg->mode & XT_HASHLIMIT_BYTES) ? skb->len : 1;
|
| 767 | dh->rateinfo.current_rate += cost;
|
| 768 |
|
| 769 | if (!dh->rateinfo.prev_window &&
|
| 770 | (dh->rateinfo.current_rate <= dh->rateinfo.burst)) {
|
| 771 | spin_unlock(&dh->lock);
|
| 772 | local_bh_enable();
|
| 773 | return !(cfg->mode & XT_HASHLIMIT_INVERT);
|
| 774 | } else {
|
| 775 | goto overlimit;
|
| 776 | }
|
| 777 | }
|
| 778 |
|
| 779 | if (cfg->mode & XT_HASHLIMIT_BYTES)
|
| 780 | cost = hashlimit_byte_cost(skb->len, dh);
|
| 781 | else
|
| 782 | cost = dh->rateinfo.cost;
|
| 783 |
|
| 784 | if (dh->rateinfo.credit >= cost) {
|
| 785 |
|
| 786 | dh->rateinfo.credit -= cost;
|
| 787 | spin_unlock(&dh->lock);
|
| 788 | local_bh_enable();
|
| 789 | return !(cfg->mode & XT_HASHLIMIT_INVERT);
|
| 790 | }
|
| 791 |
|
| 792 | overlimit:
|
| 793 | spin_unlock(&dh->lock);
|
| 794 | local_bh_enable();
|
| 795 |
|
| 796 | return cfg->mode & XT_HASHLIMIT_INVERT;
|
| 797 |
|
| 798 | hotdrop:
|
| 799 | par->hotdrop = true;
|
| 800 | return false;
|
| 801 | }
|
| 802 |
|
| 803 | static bool
|
| 804 | hashlimit_mt_v1(const struct sk_buff *skb, struct xt_action_param *par)
|
| 805 | {
|
| 806 | const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
|
| 807 | struct xt_hashlimit_htable *hinfo = info->hinfo;
|
| 808 | struct hashlimit_cfg3 cfg = {};
|
| 809 | int ret;
|
| 810 |
|
| 811 | ret = cfg_copy(&cfg, (void *)&info->cfg, 1);
|
| 812 | if (ret)
|
| 813 | return ret;
|
| 814 |
|
| 815 | return hashlimit_mt_common(skb, par, hinfo, &cfg, 1);
|
| 816 | }
|
| 817 |
|
| 818 | static bool
|
| 819 | hashlimit_mt_v2(const struct sk_buff *skb, struct xt_action_param *par)
|
| 820 | {
|
| 821 | const struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
|
| 822 | struct xt_hashlimit_htable *hinfo = info->hinfo;
|
| 823 | struct hashlimit_cfg3 cfg = {};
|
| 824 | int ret;
|
| 825 |
|
| 826 | ret = cfg_copy(&cfg, (void *)&info->cfg, 2);
|
| 827 | if (ret)
|
| 828 | return ret;
|
| 829 |
|
| 830 | return hashlimit_mt_common(skb, par, hinfo, &cfg, 2);
|
| 831 | }
|
| 832 |
|
| 833 | static bool
|
| 834 | hashlimit_mt(const struct sk_buff *skb, struct xt_action_param *par)
|
| 835 | {
|
| 836 | const struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
|
| 837 | struct xt_hashlimit_htable *hinfo = info->hinfo;
|
| 838 |
|
| 839 | return hashlimit_mt_common(skb, par, hinfo, &info->cfg, 3);
|
| 840 | }
|
| 841 |
|
| 842 | #define HASHLIMIT_MAX_SIZE 1048576
|
| 843 |
|
| 844 | static int hashlimit_mt_check_common(const struct xt_mtchk_param *par,
|
| 845 | struct xt_hashlimit_htable **hinfo,
|
| 846 | struct hashlimit_cfg3 *cfg,
|
| 847 | const char *name, int revision)
|
| 848 | {
|
| 849 | struct net *net = par->net;
|
| 850 | int ret;
|
| 851 |
|
| 852 | if (cfg->gc_interval == 0 || cfg->expire == 0)
|
| 853 | return -EINVAL;
|
| 854 | if (cfg->size > HASHLIMIT_MAX_SIZE) {
|
| 855 | cfg->size = HASHLIMIT_MAX_SIZE;
|
| 856 | pr_info_ratelimited("size too large, truncated to %u\n", cfg->size);
|
| 857 | }
|
| 858 | if (cfg->max > HASHLIMIT_MAX_SIZE) {
|
| 859 | cfg->max = HASHLIMIT_MAX_SIZE;
|
| 860 | pr_info_ratelimited("max too large, truncated to %u\n", cfg->max);
|
| 861 | }
|
| 862 | if (par->family == NFPROTO_IPV4) {
|
| 863 | if (cfg->srcmask > 32 || cfg->dstmask > 32)
|
| 864 | return -EINVAL;
|
| 865 | } else {
|
| 866 | if (cfg->srcmask > 128 || cfg->dstmask > 128)
|
| 867 | return -EINVAL;
|
| 868 | }
|
| 869 |
|
| 870 | if (cfg->mode & ~XT_HASHLIMIT_ALL) {
|
| 871 | pr_info_ratelimited("Unknown mode mask %X, kernel too old?\n",
|
| 872 | cfg->mode);
|
| 873 | return -EINVAL;
|
| 874 | }
|
| 875 |
|
| 876 |
|
| 877 | if (cfg->mode & XT_HASHLIMIT_RATE_MATCH) {
|
| 878 | if (revision < 3)
|
| 879 | return -EINVAL;
|
| 880 |
|
| 881 | if (cfg->avg == 0 || cfg->avg > U32_MAX) {
|
| 882 | pr_info_ratelimited("invalid rate\n");
|
| 883 | return -ERANGE;
|
| 884 | }
|
| 885 |
|
| 886 | if (cfg->interval == 0) {
|
| 887 | pr_info_ratelimited("invalid interval\n");
|
| 888 | return -EINVAL;
|
| 889 | }
|
| 890 | } else if (cfg->mode & XT_HASHLIMIT_BYTES) {
|
| 891 | if (user2credits_byte(cfg->avg) == 0) {
|
| 892 | pr_info_ratelimited("overflow, rate too high: %llu\n",
|
| 893 | cfg->avg);
|
| 894 | return -EINVAL;
|
| 895 | }
|
| 896 | } else if (cfg->burst == 0 ||
|
| 897 | user2credits(cfg->avg * cfg->burst, revision) <
|
| 898 | user2credits(cfg->avg, revision)) {
|
| 899 | pr_info_ratelimited("overflow, try lower: %llu/%llu\n",
|
| 900 | cfg->avg, cfg->burst);
|
| 901 | return -ERANGE;
|
| 902 | }
|
| 903 |
|
| 904 | mutex_lock(&hashlimit_mutex);
|
| 905 | *hinfo = htable_find_get(net, name, par->family);
|
| 906 | if (*hinfo == NULL) {
|
| 907 | ret = htable_create(net, cfg, name, par->family,
|
| 908 | hinfo, revision);
|
| 909 | if (ret < 0) {
|
| 910 | mutex_unlock(&hashlimit_mutex);
|
| 911 | return ret;
|
| 912 | }
|
| 913 | } else {
|
| 914 | if ((cfg->mode & XT_HASHLIMIT_RATE_MATCH &&
|
| 915 | !(*hinfo)->ratematch) ||
|
| 916 | (!(cfg->mode & XT_HASHLIMIT_RATE_MATCH) &&
|
| 917 | (*hinfo)->ratematch)) {
|
| 918 | mutex_unlock(&hashlimit_mutex);
|
| 919 | htable_put(*hinfo);
|
| 920 | return -EINVAL;
|
| 921 | }
|
| 922 | }
|
| 923 | mutex_unlock(&hashlimit_mutex);
|
| 924 |
|
| 925 | return 0;
|
| 926 | }
|
| 927 |
|
| 928 | static int hashlimit_mt_check_v1(const struct xt_mtchk_param *par)
|
| 929 | {
|
| 930 | struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
|
| 931 | struct hashlimit_cfg3 cfg = {};
|
| 932 | int ret;
|
| 933 |
|
| 934 | ret = xt_check_proc_name(info->name, sizeof(info->name));
|
| 935 | if (ret)
|
| 936 | return ret;
|
| 937 |
|
| 938 | ret = cfg_copy(&cfg, (void *)&info->cfg, 1);
|
| 939 | if (ret)
|
| 940 | return ret;
|
| 941 |
|
| 942 | return hashlimit_mt_check_common(par, &info->hinfo,
|
| 943 | &cfg, info->name, 1);
|
| 944 | }
|
| 945 |
|
| 946 | static int hashlimit_mt_check_v2(const struct xt_mtchk_param *par)
|
| 947 | {
|
| 948 | struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
|
| 949 | struct hashlimit_cfg3 cfg = {};
|
| 950 | int ret;
|
| 951 |
|
| 952 | ret = xt_check_proc_name(info->name, sizeof(info->name));
|
| 953 | if (ret)
|
| 954 | return ret;
|
| 955 |
|
| 956 | ret = cfg_copy(&cfg, (void *)&info->cfg, 2);
|
| 957 | if (ret)
|
| 958 | return ret;
|
| 959 |
|
| 960 | return hashlimit_mt_check_common(par, &info->hinfo,
|
| 961 | &cfg, info->name, 2);
|
| 962 | }
|
| 963 |
|
| 964 | static int hashlimit_mt_check(const struct xt_mtchk_param *par)
|
| 965 | {
|
| 966 | struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
|
| 967 | int ret;
|
| 968 |
|
| 969 | ret = xt_check_proc_name(info->name, sizeof(info->name));
|
| 970 | if (ret)
|
| 971 | return ret;
|
| 972 |
|
| 973 | return hashlimit_mt_check_common(par, &info->hinfo, &info->cfg,
|
| 974 | info->name, 3);
|
| 975 | }
|
| 976 |
|
| 977 | static void hashlimit_mt_destroy_v2(const struct xt_mtdtor_param *par)
|
| 978 | {
|
| 979 | const struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
|
| 980 |
|
| 981 | htable_put(info->hinfo);
|
| 982 | }
|
| 983 |
|
| 984 | static void hashlimit_mt_destroy_v1(const struct xt_mtdtor_param *par)
|
| 985 | {
|
| 986 | const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
|
| 987 |
|
| 988 | htable_put(info->hinfo);
|
| 989 | }
|
| 990 |
|
| 991 | static void hashlimit_mt_destroy(const struct xt_mtdtor_param *par)
|
| 992 | {
|
| 993 | const struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
|
| 994 |
|
| 995 | htable_put(info->hinfo);
|
| 996 | }
|
| 997 |
|
| 998 | static struct xt_match hashlimit_mt_reg[] __read_mostly = {
|
| 999 | {
|
| 1000 | .name = "hashlimit",
|
| 1001 | .revision = 1,
|
| 1002 | .family = NFPROTO_IPV4,
|
| 1003 | .match = hashlimit_mt_v1,
|
| 1004 | .matchsize = sizeof(struct xt_hashlimit_mtinfo1),
|
| 1005 | .usersize = offsetof(struct xt_hashlimit_mtinfo1, hinfo),
|
| 1006 | .checkentry = hashlimit_mt_check_v1,
|
| 1007 | .destroy = hashlimit_mt_destroy_v1,
|
| 1008 | .me = THIS_MODULE,
|
| 1009 | },
|
| 1010 | {
|
| 1011 | .name = "hashlimit",
|
| 1012 | .revision = 2,
|
| 1013 | .family = NFPROTO_IPV4,
|
| 1014 | .match = hashlimit_mt_v2,
|
| 1015 | .matchsize = sizeof(struct xt_hashlimit_mtinfo2),
|
| 1016 | .usersize = offsetof(struct xt_hashlimit_mtinfo2, hinfo),
|
| 1017 | .checkentry = hashlimit_mt_check_v2,
|
| 1018 | .destroy = hashlimit_mt_destroy_v2,
|
| 1019 | .me = THIS_MODULE,
|
| 1020 | },
|
| 1021 | {
|
| 1022 | .name = "hashlimit",
|
| 1023 | .revision = 3,
|
| 1024 | .family = NFPROTO_IPV4,
|
| 1025 | .match = hashlimit_mt,
|
| 1026 | .matchsize = sizeof(struct xt_hashlimit_mtinfo3),
|
| 1027 | .usersize = offsetof(struct xt_hashlimit_mtinfo3, hinfo),
|
| 1028 | .checkentry = hashlimit_mt_check,
|
| 1029 | .destroy = hashlimit_mt_destroy,
|
| 1030 | .me = THIS_MODULE,
|
| 1031 | },
|
| 1032 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
|
| 1033 | {
|
| 1034 | .name = "hashlimit",
|
| 1035 | .revision = 1,
|
| 1036 | .family = NFPROTO_IPV6,
|
| 1037 | .match = hashlimit_mt_v1,
|
| 1038 | .matchsize = sizeof(struct xt_hashlimit_mtinfo1),
|
| 1039 | .usersize = offsetof(struct xt_hashlimit_mtinfo1, hinfo),
|
| 1040 | .checkentry = hashlimit_mt_check_v1,
|
| 1041 | .destroy = hashlimit_mt_destroy_v1,
|
| 1042 | .me = THIS_MODULE,
|
| 1043 | },
|
| 1044 | {
|
| 1045 | .name = "hashlimit",
|
| 1046 | .revision = 2,
|
| 1047 | .family = NFPROTO_IPV6,
|
| 1048 | .match = hashlimit_mt_v2,
|
| 1049 | .matchsize = sizeof(struct xt_hashlimit_mtinfo2),
|
| 1050 | .usersize = offsetof(struct xt_hashlimit_mtinfo2, hinfo),
|
| 1051 | .checkentry = hashlimit_mt_check_v2,
|
| 1052 | .destroy = hashlimit_mt_destroy_v2,
|
| 1053 | .me = THIS_MODULE,
|
| 1054 | },
|
| 1055 | {
|
| 1056 | .name = "hashlimit",
|
| 1057 | .revision = 3,
|
| 1058 | .family = NFPROTO_IPV6,
|
| 1059 | .match = hashlimit_mt,
|
| 1060 | .matchsize = sizeof(struct xt_hashlimit_mtinfo3),
|
| 1061 | .usersize = offsetof(struct xt_hashlimit_mtinfo3, hinfo),
|
| 1062 | .checkentry = hashlimit_mt_check,
|
| 1063 | .destroy = hashlimit_mt_destroy,
|
| 1064 | .me = THIS_MODULE,
|
| 1065 | },
|
| 1066 | #endif
|
| 1067 | };
|
| 1068 |
|
| 1069 |
|
| 1070 | static void *dl_seq_start(struct seq_file *s, loff_t *pos)
|
| 1071 | __acquires(htable->lock)
|
| 1072 | {
|
| 1073 | struct xt_hashlimit_htable *htable = pde_data(file_inode(s->file));
|
| 1074 | unsigned int *bucket;
|
| 1075 |
|
| 1076 | spin_lock_bh(&htable->lock);
|
| 1077 | if (*pos >= htable->cfg.size)
|
| 1078 | return NULL;
|
| 1079 |
|
| 1080 | bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC);
|
| 1081 | if (!bucket)
|
| 1082 | return ERR_PTR(-ENOMEM);
|
| 1083 |
|
| 1084 | *bucket = *pos;
|
| 1085 | return bucket;
|
| 1086 | }
|
| 1087 |
|
| 1088 | static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
|
| 1089 | {
|
| 1090 | struct xt_hashlimit_htable *htable = pde_data(file_inode(s->file));
|
| 1091 | unsigned int *bucket = v;
|
| 1092 |
|
| 1093 | *pos = ++(*bucket);
|
| 1094 | if (*pos >= htable->cfg.size) {
|
| 1095 | kfree(v);
|
| 1096 | return NULL;
|
| 1097 | }
|
| 1098 | return bucket;
|
| 1099 | }
|
| 1100 |
|
| 1101 | static void dl_seq_stop(struct seq_file *s, void *v)
|
| 1102 | __releases(htable->lock)
|
| 1103 | {
|
| 1104 | struct xt_hashlimit_htable *htable = pde_data(file_inode(s->file));
|
| 1105 | unsigned int *bucket = v;
|
| 1106 |
|
| 1107 | if (!IS_ERR(bucket))
|
| 1108 | kfree(bucket);
|
| 1109 | spin_unlock_bh(&htable->lock);
|
| 1110 | }
|
| 1111 |
|
| 1112 | static void dl_seq_print(struct dsthash_ent *ent, u_int8_t family,
|
| 1113 | struct seq_file *s)
|
| 1114 | {
|
| 1115 | switch (family) {
|
| 1116 | case NFPROTO_IPV4:
|
| 1117 | seq_printf(s, "%ld %pI4:%u->%pI4:%u %llu %llu %llu\n",
|
| 1118 | (long)(ent->expires - jiffies)/HZ,
|
| 1119 | &ent->dst.ip.src,
|
| 1120 | ntohs(ent->dst.src_port),
|
| 1121 | &ent->dst.ip.dst,
|
| 1122 | ntohs(ent->dst.dst_port),
|
| 1123 | ent->rateinfo.credit, ent->rateinfo.credit_cap,
|
| 1124 | ent->rateinfo.cost);
|
| 1125 | break;
|
| 1126 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
|
| 1127 | case NFPROTO_IPV6:
|
| 1128 | seq_printf(s, "%ld %pI6:%u->%pI6:%u %llu %llu %llu\n",
|
| 1129 | (long)(ent->expires - jiffies)/HZ,
|
| 1130 | &ent->dst.ip6.src,
|
| 1131 | ntohs(ent->dst.src_port),
|
| 1132 | &ent->dst.ip6.dst,
|
| 1133 | ntohs(ent->dst.dst_port),
|
| 1134 | ent->rateinfo.credit, ent->rateinfo.credit_cap,
|
| 1135 | ent->rateinfo.cost);
|
| 1136 | break;
|
| 1137 | #endif
|
| 1138 | default:
|
| 1139 | BUG();
|
| 1140 | }
|
| 1141 | }
|
| 1142 |
|
| 1143 | static int dl_seq_real_show_v2(struct dsthash_ent *ent, u_int8_t family,
|
| 1144 | struct seq_file *s)
|
| 1145 | {
|
| 1146 | struct xt_hashlimit_htable *ht = pde_data(file_inode(s->file));
|
| 1147 |
|
| 1148 | spin_lock(&ent->lock);
|
| 1149 |
|
| 1150 | rateinfo_recalc(ent, jiffies, ht->cfg.mode, 2);
|
| 1151 |
|
| 1152 | dl_seq_print(ent, family, s);
|
| 1153 |
|
| 1154 | spin_unlock(&ent->lock);
|
| 1155 | return seq_has_overflowed(s);
|
| 1156 | }
|
| 1157 |
|
| 1158 | static int dl_seq_real_show_v1(struct dsthash_ent *ent, u_int8_t family,
|
| 1159 | struct seq_file *s)
|
| 1160 | {
|
| 1161 | struct xt_hashlimit_htable *ht = pde_data(file_inode(s->file));
|
| 1162 |
|
| 1163 | spin_lock(&ent->lock);
|
| 1164 |
|
| 1165 | rateinfo_recalc(ent, jiffies, ht->cfg.mode, 1);
|
| 1166 |
|
| 1167 | dl_seq_print(ent, family, s);
|
| 1168 |
|
| 1169 | spin_unlock(&ent->lock);
|
| 1170 | return seq_has_overflowed(s);
|
| 1171 | }
|
| 1172 |
|
| 1173 | static int dl_seq_real_show(struct dsthash_ent *ent, u_int8_t family,
|
| 1174 | struct seq_file *s)
|
| 1175 | {
|
| 1176 | struct xt_hashlimit_htable *ht = pde_data(file_inode(s->file));
|
| 1177 |
|
| 1178 | spin_lock(&ent->lock);
|
| 1179 |
|
| 1180 | rateinfo_recalc(ent, jiffies, ht->cfg.mode, 3);
|
| 1181 |
|
| 1182 | dl_seq_print(ent, family, s);
|
| 1183 |
|
| 1184 | spin_unlock(&ent->lock);
|
| 1185 | return seq_has_overflowed(s);
|
| 1186 | }
|
| 1187 |
|
| 1188 | static int dl_seq_show_v2(struct seq_file *s, void *v)
|
| 1189 | {
|
| 1190 | struct xt_hashlimit_htable *htable = pde_data(file_inode(s->file));
|
| 1191 | unsigned int *bucket = (unsigned int *)v;
|
| 1192 | struct dsthash_ent *ent;
|
| 1193 |
|
| 1194 | if (!hlist_empty(&htable->hash[*bucket])) {
|
| 1195 | hlist_for_each_entry(ent, &htable->hash[*bucket], node)
|
| 1196 | if (dl_seq_real_show_v2(ent, htable->family, s))
|
| 1197 | return -1;
|
| 1198 | }
|
| 1199 | return 0;
|
| 1200 | }
|
| 1201 |
|
| 1202 | static int dl_seq_show_v1(struct seq_file *s, void *v)
|
| 1203 | {
|
| 1204 | struct xt_hashlimit_htable *htable = pde_data(file_inode(s->file));
|
| 1205 | unsigned int *bucket = v;
|
| 1206 | struct dsthash_ent *ent;
|
| 1207 |
|
| 1208 | if (!hlist_empty(&htable->hash[*bucket])) {
|
| 1209 | hlist_for_each_entry(ent, &htable->hash[*bucket], node)
|
| 1210 | if (dl_seq_real_show_v1(ent, htable->family, s))
|
| 1211 | return -1;
|
| 1212 | }
|
| 1213 | return 0;
|
| 1214 | }
|
| 1215 |
|
| 1216 | static int dl_seq_show(struct seq_file *s, void *v)
|
| 1217 | {
|
| 1218 | struct xt_hashlimit_htable *htable = pde_data(file_inode(s->file));
|
| 1219 | unsigned int *bucket = v;
|
| 1220 | struct dsthash_ent *ent;
|
| 1221 |
|
| 1222 | if (!hlist_empty(&htable->hash[*bucket])) {
|
| 1223 | hlist_for_each_entry(ent, &htable->hash[*bucket], node)
|
| 1224 | if (dl_seq_real_show(ent, htable->family, s))
|
| 1225 | return -1;
|
| 1226 | }
|
| 1227 | return 0;
|
| 1228 | }
|
| 1229 |
|
| 1230 | static const struct seq_operations dl_seq_ops_v1 = {
|
| 1231 | .start = dl_seq_start,
|
| 1232 | .next = dl_seq_next,
|
| 1233 | .stop = dl_seq_stop,
|
| 1234 | .show = dl_seq_show_v1
|
| 1235 | };
|
| 1236 |
|
| 1237 | static const struct seq_operations dl_seq_ops_v2 = {
|
| 1238 | .start = dl_seq_start,
|
| 1239 | .next = dl_seq_next,
|
| 1240 | .stop = dl_seq_stop,
|
| 1241 | .show = dl_seq_show_v2
|
| 1242 | };
|
| 1243 |
|
| 1244 | static const struct seq_operations dl_seq_ops = {
|
| 1245 | .start = dl_seq_start,
|
| 1246 | .next = dl_seq_next,
|
| 1247 | .stop = dl_seq_stop,
|
| 1248 | .show = dl_seq_show
|
| 1249 | };
|
| 1250 |
|
| 1251 | static int __net_init hashlimit_proc_net_init(struct net *net)
|
| 1252 | {
|
| 1253 | struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
|
| 1254 |
|
| 1255 | hashlimit_net->ipt_hashlimit = proc_mkdir("ipt_hashlimit", net->proc_net);
|
| 1256 | if (!hashlimit_net->ipt_hashlimit)
|
| 1257 | return -ENOMEM;
|
| 1258 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
|
| 1259 | hashlimit_net->ip6t_hashlimit = proc_mkdir("ip6t_hashlimit", net->proc_net);
|
| 1260 | if (!hashlimit_net->ip6t_hashlimit) {
|
| 1261 | remove_proc_entry("ipt_hashlimit", net->proc_net);
|
| 1262 | return -ENOMEM;
|
| 1263 | }
|
| 1264 | #endif
|
| 1265 | return 0;
|
| 1266 | }
|
| 1267 |
|
| 1268 | static void __net_exit hashlimit_proc_net_exit(struct net *net)
|
| 1269 | {
|
| 1270 | struct xt_hashlimit_htable *hinfo;
|
| 1271 | struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
|
| 1272 |
|
| 1273 | |
| 1274 | * Make sure that the parent ipt_hashlimit and ip6t_hashlimit proc
|
| 1275 | * entries is empty before trying to remove it.
|
| 1276 |
|
| 1277 | mutex_lock(&hashlimit_mutex);
|
| 1278 | hlist_for_each_entry(hinfo, &hashlimit_net->htables, node)
|
| 1279 | htable_remove_proc_entry(hinfo);
|
| 1280 | hashlimit_net->ipt_hashlimit = NULL;
|
| 1281 | hashlimit_net->ip6t_hashlimit = NULL;
|
| 1282 | mutex_unlock(&hashlimit_mutex);
|
| 1283 |
|
| 1284 | remove_proc_entry("ipt_hashlimit", net->proc_net);
|
| 1285 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
|
| 1286 | remove_proc_entry("ip6t_hashlimit", net->proc_net);
|
| 1287 | #endif
|
| 1288 | }
|
| 1289 |
|
| 1290 | static int __net_init hashlimit_net_init(struct net *net)
|
| 1291 | {
|
| 1292 | struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
|
| 1293 |
|
| 1294 | INIT_HLIST_HEAD(&hashlimit_net->htables);
|
| 1295 | return hashlimit_proc_net_init(net);
|
| 1296 | }
|
| 1297 |
|
| 1298 | static void __net_exit hashlimit_net_exit(struct net *net)
|
| 1299 | {
|
| 1300 | hashlimit_proc_net_exit(net);
|
| 1301 | }
|
| 1302 |
|
| 1303 | static struct pernet_operations hashlimit_net_ops = {
|
| 1304 | .init = hashlimit_net_init,
|
| 1305 | .exit = hashlimit_net_exit,
|
| 1306 | .id = &hashlimit_net_id,
|
| 1307 | .size = sizeof(struct hashlimit_net),
|
| 1308 | };
|
| 1309 |
|
| 1310 | static int __init hashlimit_mt_init(void)
|
| 1311 | {
|
| 1312 | int err;
|
| 1313 |
|
| 1314 | err = register_pernet_subsys(&hashlimit_net_ops);
|
| 1315 | if (err < 0)
|
| 1316 | return err;
|
| 1317 | err = xt_register_matches(hashlimit_mt_reg,
|
| 1318 | ARRAY_SIZE(hashlimit_mt_reg));
|
| 1319 | if (err < 0)
|
| 1320 | goto err1;
|
| 1321 |
|
| 1322 | err = -ENOMEM;
|
| 1323 | hashlimit_cachep = kmem_cache_create("xt_hashlimit",
|
| 1324 | sizeof(struct dsthash_ent), 0, 0,
|
| 1325 | NULL);
|
| 1326 | if (!hashlimit_cachep) {
|
| 1327 | pr_warn("unable to create slab cache\n");
|
| 1328 | goto err2;
|
| 1329 | }
|
| 1330 | return 0;
|
| 1331 |
|
| 1332 | err2:
|
| 1333 | xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
|
| 1334 | err1:
|
| 1335 | unregister_pernet_subsys(&hashlimit_net_ops);
|
| 1336 | return err;
|
| 1337 |
|
| 1338 | }
|
| 1339 |
|
| 1340 | static void __exit hashlimit_mt_exit(void)
|
| 1341 | {
|
| 1342 | xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
|
| 1343 | unregister_pernet_subsys(&hashlimit_net_ops);
|
| 1344 |
|
| 1345 | rcu_barrier();
|
| 1346 | kmem_cache_destroy(hashlimit_cachep);
|
| 1347 | }
|
| 1348 |
|
| 1349 | module_init(hashlimit_mt_init);
|
| 1350 | module_exit(hashlimit_mt_exit);
|