| 1 | // SPDX-License-Identifier: GPL-2.0-only |
| 2 | /* |
| 3 | * Copyright (c) 2008-2009 Patrick McHardy <kaber@trash.net> |
| 4 | * |
| 5 | * Development of this code funded by Astaro AG (http://www.astaro.com/) |
| 6 | */ |
| 7 | |
| 8 | #include <linux/kernel.h> |
| 9 | #include <linux/init.h> |
| 10 | #include <linux/module.h> |
| 11 | #include <linux/spinlock.h> |
| 12 | #include <linux/netlink.h> |
| 13 | #include <linux/netfilter.h> |
| 14 | #include <linux/netfilter/nf_tables.h> |
| 15 | #include <net/netfilter/nf_tables.h> |
| 16 | |
| 17 | struct nft_limit { |
| 18 | spinlock_t lock; |
| 19 | u64 last; |
| 20 | u64 tokens; |
| 21 | }; |
| 22 | |
| 23 | struct nft_limit_priv { |
| 24 | struct nft_limit *limit; |
| 25 | u64 tokens_max; |
| 26 | u64 rate; |
| 27 | u64 nsecs; |
| 28 | u32 burst; |
| 29 | bool invert; |
| 30 | }; |
| 31 | |
| 32 | static inline bool nft_limit_eval(struct nft_limit_priv *priv, u64 cost) |
| 33 | { |
| 34 | u64 now, tokens; |
| 35 | s64 delta; |
| 36 | |
| 37 | spin_lock_bh(lock: &priv->limit->lock); |
| 38 | now = ktime_get_ns(); |
| 39 | tokens = priv->limit->tokens + now - priv->limit->last; |
| 40 | if (tokens > priv->tokens_max) |
| 41 | tokens = priv->tokens_max; |
| 42 | |
| 43 | priv->limit->last = now; |
| 44 | delta = tokens - cost; |
| 45 | if (delta >= 0) { |
| 46 | priv->limit->tokens = delta; |
| 47 | spin_unlock_bh(lock: &priv->limit->lock); |
| 48 | return priv->invert; |
| 49 | } |
| 50 | priv->limit->tokens = tokens; |
| 51 | spin_unlock_bh(lock: &priv->limit->lock); |
| 52 | return !priv->invert; |
| 53 | } |
| 54 | |
| 55 | /* Use same default as in iptables. */ |
| 56 | #define NFT_LIMIT_PKT_BURST_DEFAULT 5 |
| 57 | |
| 58 | static int nft_limit_init(struct nft_limit_priv *priv, |
| 59 | const struct nlattr * const tb[], bool pkts) |
| 60 | { |
| 61 | u64 unit, tokens, rate_with_burst; |
| 62 | bool invert = false; |
| 63 | |
| 64 | if (tb[NFTA_LIMIT_RATE] == NULL || |
| 65 | tb[NFTA_LIMIT_UNIT] == NULL) |
| 66 | return -EINVAL; |
| 67 | |
| 68 | priv->rate = be64_to_cpu(nla_get_be64(tb[NFTA_LIMIT_RATE])); |
| 69 | if (priv->rate == 0) |
| 70 | return -EINVAL; |
| 71 | |
| 72 | unit = be64_to_cpu(nla_get_be64(tb[NFTA_LIMIT_UNIT])); |
| 73 | if (check_mul_overflow(unit, NSEC_PER_SEC, &priv->nsecs)) |
| 74 | return -EOVERFLOW; |
| 75 | |
| 76 | if (tb[NFTA_LIMIT_BURST]) |
| 77 | priv->burst = ntohl(nla_get_be32(tb[NFTA_LIMIT_BURST])); |
| 78 | |
| 79 | if (pkts && priv->burst == 0) |
| 80 | priv->burst = NFT_LIMIT_PKT_BURST_DEFAULT; |
| 81 | |
| 82 | if (check_add_overflow(priv->rate, priv->burst, &rate_with_burst)) |
| 83 | return -EOVERFLOW; |
| 84 | |
| 85 | if (pkts) { |
| 86 | u64 tmp = div64_u64(dividend: priv->nsecs, divisor: priv->rate); |
| 87 | |
| 88 | if (check_mul_overflow(tmp, priv->burst, &tokens)) |
| 89 | return -EOVERFLOW; |
| 90 | } else { |
| 91 | u64 tmp; |
| 92 | |
| 93 | /* The token bucket size limits the number of tokens can be |
| 94 | * accumulated. tokens_max specifies the bucket size. |
| 95 | * tokens_max = unit * (rate + burst) / rate. |
| 96 | */ |
| 97 | if (check_mul_overflow(priv->nsecs, rate_with_burst, &tmp)) |
| 98 | return -EOVERFLOW; |
| 99 | |
| 100 | tokens = div64_u64(dividend: tmp, divisor: priv->rate); |
| 101 | } |
| 102 | |
| 103 | if (tb[NFTA_LIMIT_FLAGS]) { |
| 104 | u32 flags = ntohl(nla_get_be32(tb[NFTA_LIMIT_FLAGS])); |
| 105 | |
| 106 | if (flags & ~NFT_LIMIT_F_INV) |
| 107 | return -EOPNOTSUPP; |
| 108 | |
| 109 | if (flags & NFT_LIMIT_F_INV) |
| 110 | invert = true; |
| 111 | } |
| 112 | |
| 113 | priv->limit = kmalloc(sizeof(*priv->limit), GFP_KERNEL_ACCOUNT); |
| 114 | if (!priv->limit) |
| 115 | return -ENOMEM; |
| 116 | |
| 117 | priv->limit->tokens = tokens; |
| 118 | priv->tokens_max = priv->limit->tokens; |
| 119 | priv->invert = invert; |
| 120 | priv->limit->last = ktime_get_ns(); |
| 121 | spin_lock_init(&priv->limit->lock); |
| 122 | |
| 123 | return 0; |
| 124 | } |
| 125 | |
| 126 | static int nft_limit_dump(struct sk_buff *skb, const struct nft_limit_priv *priv, |
| 127 | enum nft_limit_type type) |
| 128 | { |
| 129 | u32 flags = priv->invert ? NFT_LIMIT_F_INV : 0; |
| 130 | u64 secs = div_u64(dividend: priv->nsecs, NSEC_PER_SEC); |
| 131 | |
| 132 | if (nla_put_be64(skb, attrtype: NFTA_LIMIT_RATE, cpu_to_be64(priv->rate), |
| 133 | padattr: NFTA_LIMIT_PAD) || |
| 134 | nla_put_be64(skb, attrtype: NFTA_LIMIT_UNIT, cpu_to_be64(secs), |
| 135 | padattr: NFTA_LIMIT_PAD) || |
| 136 | nla_put_be32(skb, attrtype: NFTA_LIMIT_BURST, htonl(priv->burst)) || |
| 137 | nla_put_be32(skb, attrtype: NFTA_LIMIT_TYPE, htonl(type)) || |
| 138 | nla_put_be32(skb, attrtype: NFTA_LIMIT_FLAGS, htonl(flags))) |
| 139 | goto nla_put_failure; |
| 140 | return 0; |
| 141 | |
| 142 | nla_put_failure: |
| 143 | return -1; |
| 144 | } |
| 145 | |
| 146 | static void nft_limit_destroy(const struct nft_ctx *ctx, |
| 147 | const struct nft_limit_priv *priv) |
| 148 | { |
| 149 | kfree(objp: priv->limit); |
| 150 | } |
| 151 | |
| 152 | static int nft_limit_clone(struct nft_limit_priv *priv_dst, |
| 153 | const struct nft_limit_priv *priv_src, gfp_t gfp) |
| 154 | { |
| 155 | priv_dst->tokens_max = priv_src->tokens_max; |
| 156 | priv_dst->rate = priv_src->rate; |
| 157 | priv_dst->nsecs = priv_src->nsecs; |
| 158 | priv_dst->burst = priv_src->burst; |
| 159 | priv_dst->invert = priv_src->invert; |
| 160 | |
| 161 | priv_dst->limit = kmalloc(sizeof(*priv_dst->limit), gfp); |
| 162 | if (!priv_dst->limit) |
| 163 | return -ENOMEM; |
| 164 | |
| 165 | spin_lock_init(&priv_dst->limit->lock); |
| 166 | priv_dst->limit->tokens = priv_src->tokens_max; |
| 167 | priv_dst->limit->last = ktime_get_ns(); |
| 168 | |
| 169 | return 0; |
| 170 | } |
| 171 | |
| 172 | struct nft_limit_priv_pkts { |
| 173 | struct nft_limit_priv limit; |
| 174 | u64 cost; |
| 175 | }; |
| 176 | |
| 177 | static void nft_limit_pkts_eval(const struct nft_expr *expr, |
| 178 | struct nft_regs *regs, |
| 179 | const struct nft_pktinfo *pkt) |
| 180 | { |
| 181 | struct nft_limit_priv_pkts *priv = nft_expr_priv(expr); |
| 182 | |
| 183 | if (nft_limit_eval(priv: &priv->limit, cost: priv->cost)) |
| 184 | regs->verdict.code = NFT_BREAK; |
| 185 | } |
| 186 | |
| 187 | static const struct nla_policy nft_limit_policy[NFTA_LIMIT_MAX + 1] = { |
| 188 | [NFTA_LIMIT_RATE] = { .type = NLA_U64 }, |
| 189 | [NFTA_LIMIT_UNIT] = { .type = NLA_U64 }, |
| 190 | [NFTA_LIMIT_BURST] = { .type = NLA_U32 }, |
| 191 | [NFTA_LIMIT_TYPE] = { .type = NLA_U32 }, |
| 192 | [NFTA_LIMIT_FLAGS] = { .type = NLA_U32 }, |
| 193 | }; |
| 194 | |
| 195 | static int nft_limit_pkts_init(const struct nft_ctx *ctx, |
| 196 | const struct nft_expr *expr, |
| 197 | const struct nlattr * const tb[]) |
| 198 | { |
| 199 | struct nft_limit_priv_pkts *priv = nft_expr_priv(expr); |
| 200 | int err; |
| 201 | |
| 202 | err = nft_limit_init(priv: &priv->limit, tb, pkts: true); |
| 203 | if (err < 0) |
| 204 | return err; |
| 205 | |
| 206 | priv->cost = div64_u64(dividend: priv->limit.nsecs, divisor: priv->limit.rate); |
| 207 | return 0; |
| 208 | } |
| 209 | |
| 210 | static int nft_limit_pkts_dump(struct sk_buff *skb, |
| 211 | const struct nft_expr *expr, bool reset) |
| 212 | { |
| 213 | const struct nft_limit_priv_pkts *priv = nft_expr_priv(expr); |
| 214 | |
| 215 | return nft_limit_dump(skb, priv: &priv->limit, type: NFT_LIMIT_PKTS); |
| 216 | } |
| 217 | |
| 218 | static void nft_limit_pkts_destroy(const struct nft_ctx *ctx, |
| 219 | const struct nft_expr *expr) |
| 220 | { |
| 221 | const struct nft_limit_priv_pkts *priv = nft_expr_priv(expr); |
| 222 | |
| 223 | nft_limit_destroy(ctx, priv: &priv->limit); |
| 224 | } |
| 225 | |
| 226 | static int nft_limit_pkts_clone(struct nft_expr *dst, const struct nft_expr *src, |
| 227 | gfp_t gfp) |
| 228 | { |
| 229 | struct nft_limit_priv_pkts *priv_dst = nft_expr_priv(expr: dst); |
| 230 | struct nft_limit_priv_pkts *priv_src = nft_expr_priv(expr: src); |
| 231 | |
| 232 | priv_dst->cost = priv_src->cost; |
| 233 | |
| 234 | return nft_limit_clone(priv_dst: &priv_dst->limit, priv_src: &priv_src->limit, gfp); |
| 235 | } |
| 236 | |
| 237 | static struct nft_expr_type nft_limit_type; |
| 238 | static const struct nft_expr_ops nft_limit_pkts_ops = { |
| 239 | .type = &nft_limit_type, |
| 240 | .size = NFT_EXPR_SIZE(sizeof(struct nft_limit_priv_pkts)), |
| 241 | .eval = nft_limit_pkts_eval, |
| 242 | .init = nft_limit_pkts_init, |
| 243 | .destroy = nft_limit_pkts_destroy, |
| 244 | .clone = nft_limit_pkts_clone, |
| 245 | .dump = nft_limit_pkts_dump, |
| 246 | .reduce = NFT_REDUCE_READONLY, |
| 247 | }; |
| 248 | |
| 249 | static void nft_limit_bytes_eval(const struct nft_expr *expr, |
| 250 | struct nft_regs *regs, |
| 251 | const struct nft_pktinfo *pkt) |
| 252 | { |
| 253 | struct nft_limit_priv *priv = nft_expr_priv(expr); |
| 254 | u64 cost = div64_u64(dividend: priv->nsecs * pkt->skb->len, divisor: priv->rate); |
| 255 | |
| 256 | if (nft_limit_eval(priv, cost)) |
| 257 | regs->verdict.code = NFT_BREAK; |
| 258 | } |
| 259 | |
| 260 | static int nft_limit_bytes_init(const struct nft_ctx *ctx, |
| 261 | const struct nft_expr *expr, |
| 262 | const struct nlattr * const tb[]) |
| 263 | { |
| 264 | struct nft_limit_priv *priv = nft_expr_priv(expr); |
| 265 | |
| 266 | return nft_limit_init(priv, tb, pkts: false); |
| 267 | } |
| 268 | |
| 269 | static int nft_limit_bytes_dump(struct sk_buff *skb, |
| 270 | const struct nft_expr *expr, bool reset) |
| 271 | { |
| 272 | const struct nft_limit_priv *priv = nft_expr_priv(expr); |
| 273 | |
| 274 | return nft_limit_dump(skb, priv, type: NFT_LIMIT_PKT_BYTES); |
| 275 | } |
| 276 | |
| 277 | static void nft_limit_bytes_destroy(const struct nft_ctx *ctx, |
| 278 | const struct nft_expr *expr) |
| 279 | { |
| 280 | const struct nft_limit_priv *priv = nft_expr_priv(expr); |
| 281 | |
| 282 | nft_limit_destroy(ctx, priv); |
| 283 | } |
| 284 | |
| 285 | static int nft_limit_bytes_clone(struct nft_expr *dst, const struct nft_expr *src, |
| 286 | gfp_t gfp) |
| 287 | { |
| 288 | struct nft_limit_priv *priv_dst = nft_expr_priv(expr: dst); |
| 289 | struct nft_limit_priv *priv_src = nft_expr_priv(expr: src); |
| 290 | |
| 291 | return nft_limit_clone(priv_dst, priv_src, gfp); |
| 292 | } |
| 293 | |
| 294 | static const struct nft_expr_ops nft_limit_bytes_ops = { |
| 295 | .type = &nft_limit_type, |
| 296 | .size = NFT_EXPR_SIZE(sizeof(struct nft_limit_priv)), |
| 297 | .eval = nft_limit_bytes_eval, |
| 298 | .init = nft_limit_bytes_init, |
| 299 | .dump = nft_limit_bytes_dump, |
| 300 | .clone = nft_limit_bytes_clone, |
| 301 | .destroy = nft_limit_bytes_destroy, |
| 302 | .reduce = NFT_REDUCE_READONLY, |
| 303 | }; |
| 304 | |
| 305 | static const struct nft_expr_ops * |
| 306 | nft_limit_select_ops(const struct nft_ctx *ctx, |
| 307 | const struct nlattr * const tb[]) |
| 308 | { |
| 309 | if (tb[NFTA_LIMIT_TYPE] == NULL) |
| 310 | return &nft_limit_pkts_ops; |
| 311 | |
| 312 | switch (ntohl(nla_get_be32(tb[NFTA_LIMIT_TYPE]))) { |
| 313 | case NFT_LIMIT_PKTS: |
| 314 | return &nft_limit_pkts_ops; |
| 315 | case NFT_LIMIT_PKT_BYTES: |
| 316 | return &nft_limit_bytes_ops; |
| 317 | } |
| 318 | return ERR_PTR(error: -EOPNOTSUPP); |
| 319 | } |
| 320 | |
| 321 | static struct nft_expr_type nft_limit_type __read_mostly = { |
| 322 | .name = "limit" , |
| 323 | .select_ops = nft_limit_select_ops, |
| 324 | .policy = nft_limit_policy, |
| 325 | .maxattr = NFTA_LIMIT_MAX, |
| 326 | .flags = NFT_EXPR_STATEFUL, |
| 327 | .owner = THIS_MODULE, |
| 328 | }; |
| 329 | |
| 330 | static void nft_limit_obj_pkts_eval(struct nft_object *obj, |
| 331 | struct nft_regs *regs, |
| 332 | const struct nft_pktinfo *pkt) |
| 333 | { |
| 334 | struct nft_limit_priv_pkts *priv = nft_obj_data(obj); |
| 335 | |
| 336 | if (nft_limit_eval(priv: &priv->limit, cost: priv->cost)) |
| 337 | regs->verdict.code = NFT_BREAK; |
| 338 | } |
| 339 | |
| 340 | static int nft_limit_obj_pkts_init(const struct nft_ctx *ctx, |
| 341 | const struct nlattr * const tb[], |
| 342 | struct nft_object *obj) |
| 343 | { |
| 344 | struct nft_limit_priv_pkts *priv = nft_obj_data(obj); |
| 345 | int err; |
| 346 | |
| 347 | err = nft_limit_init(priv: &priv->limit, tb, pkts: true); |
| 348 | if (err < 0) |
| 349 | return err; |
| 350 | |
| 351 | priv->cost = div64_u64(dividend: priv->limit.nsecs, divisor: priv->limit.rate); |
| 352 | return 0; |
| 353 | } |
| 354 | |
| 355 | static int nft_limit_obj_pkts_dump(struct sk_buff *skb, |
| 356 | struct nft_object *obj, |
| 357 | bool reset) |
| 358 | { |
| 359 | const struct nft_limit_priv_pkts *priv = nft_obj_data(obj); |
| 360 | |
| 361 | return nft_limit_dump(skb, priv: &priv->limit, type: NFT_LIMIT_PKTS); |
| 362 | } |
| 363 | |
| 364 | static void nft_limit_obj_pkts_destroy(const struct nft_ctx *ctx, |
| 365 | struct nft_object *obj) |
| 366 | { |
| 367 | struct nft_limit_priv_pkts *priv = nft_obj_data(obj); |
| 368 | |
| 369 | nft_limit_destroy(ctx, priv: &priv->limit); |
| 370 | } |
| 371 | |
| 372 | static struct nft_object_type nft_limit_obj_type; |
| 373 | static const struct nft_object_ops nft_limit_obj_pkts_ops = { |
| 374 | .type = &nft_limit_obj_type, |
| 375 | .size = NFT_EXPR_SIZE(sizeof(struct nft_limit_priv_pkts)), |
| 376 | .init = nft_limit_obj_pkts_init, |
| 377 | .destroy = nft_limit_obj_pkts_destroy, |
| 378 | .eval = nft_limit_obj_pkts_eval, |
| 379 | .dump = nft_limit_obj_pkts_dump, |
| 380 | }; |
| 381 | |
| 382 | static void nft_limit_obj_bytes_eval(struct nft_object *obj, |
| 383 | struct nft_regs *regs, |
| 384 | const struct nft_pktinfo *pkt) |
| 385 | { |
| 386 | struct nft_limit_priv *priv = nft_obj_data(obj); |
| 387 | u64 cost = div64_u64(dividend: priv->nsecs * pkt->skb->len, divisor: priv->rate); |
| 388 | |
| 389 | if (nft_limit_eval(priv, cost)) |
| 390 | regs->verdict.code = NFT_BREAK; |
| 391 | } |
| 392 | |
| 393 | static int nft_limit_obj_bytes_init(const struct nft_ctx *ctx, |
| 394 | const struct nlattr * const tb[], |
| 395 | struct nft_object *obj) |
| 396 | { |
| 397 | struct nft_limit_priv *priv = nft_obj_data(obj); |
| 398 | |
| 399 | return nft_limit_init(priv, tb, pkts: false); |
| 400 | } |
| 401 | |
| 402 | static int nft_limit_obj_bytes_dump(struct sk_buff *skb, |
| 403 | struct nft_object *obj, |
| 404 | bool reset) |
| 405 | { |
| 406 | const struct nft_limit_priv *priv = nft_obj_data(obj); |
| 407 | |
| 408 | return nft_limit_dump(skb, priv, type: NFT_LIMIT_PKT_BYTES); |
| 409 | } |
| 410 | |
| 411 | static void nft_limit_obj_bytes_destroy(const struct nft_ctx *ctx, |
| 412 | struct nft_object *obj) |
| 413 | { |
| 414 | struct nft_limit_priv *priv = nft_obj_data(obj); |
| 415 | |
| 416 | nft_limit_destroy(ctx, priv); |
| 417 | } |
| 418 | |
| 419 | static struct nft_object_type nft_limit_obj_type; |
| 420 | static const struct nft_object_ops nft_limit_obj_bytes_ops = { |
| 421 | .type = &nft_limit_obj_type, |
| 422 | .size = sizeof(struct nft_limit_priv), |
| 423 | .init = nft_limit_obj_bytes_init, |
| 424 | .destroy = nft_limit_obj_bytes_destroy, |
| 425 | .eval = nft_limit_obj_bytes_eval, |
| 426 | .dump = nft_limit_obj_bytes_dump, |
| 427 | }; |
| 428 | |
| 429 | static const struct nft_object_ops * |
| 430 | nft_limit_obj_select_ops(const struct nft_ctx *ctx, |
| 431 | const struct nlattr * const tb[]) |
| 432 | { |
| 433 | if (!tb[NFTA_LIMIT_TYPE]) |
| 434 | return &nft_limit_obj_pkts_ops; |
| 435 | |
| 436 | switch (ntohl(nla_get_be32(tb[NFTA_LIMIT_TYPE]))) { |
| 437 | case NFT_LIMIT_PKTS: |
| 438 | return &nft_limit_obj_pkts_ops; |
| 439 | case NFT_LIMIT_PKT_BYTES: |
| 440 | return &nft_limit_obj_bytes_ops; |
| 441 | } |
| 442 | return ERR_PTR(error: -EOPNOTSUPP); |
| 443 | } |
| 444 | |
| 445 | static struct nft_object_type nft_limit_obj_type __read_mostly = { |
| 446 | .select_ops = nft_limit_obj_select_ops, |
| 447 | .type = NFT_OBJECT_LIMIT, |
| 448 | .maxattr = NFTA_LIMIT_MAX, |
| 449 | .policy = nft_limit_policy, |
| 450 | .owner = THIS_MODULE, |
| 451 | }; |
| 452 | |
| 453 | static int __init nft_limit_module_init(void) |
| 454 | { |
| 455 | int err; |
| 456 | |
| 457 | err = nft_register_obj(obj_type: &nft_limit_obj_type); |
| 458 | if (err < 0) |
| 459 | return err; |
| 460 | |
| 461 | err = nft_register_expr(&nft_limit_type); |
| 462 | if (err < 0) |
| 463 | goto err1; |
| 464 | |
| 465 | return 0; |
| 466 | err1: |
| 467 | nft_unregister_obj(obj_type: &nft_limit_obj_type); |
| 468 | return err; |
| 469 | } |
| 470 | |
| 471 | static void __exit nft_limit_module_exit(void) |
| 472 | { |
| 473 | nft_unregister_expr(&nft_limit_type); |
| 474 | nft_unregister_obj(obj_type: &nft_limit_obj_type); |
| 475 | } |
| 476 | |
| 477 | module_init(nft_limit_module_init); |
| 478 | module_exit(nft_limit_module_exit); |
| 479 | |
| 480 | MODULE_LICENSE("GPL" ); |
| 481 | MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>" ); |
| 482 | MODULE_ALIAS_NFT_EXPR("limit" ); |
| 483 | MODULE_ALIAS_NFT_OBJ(NFT_OBJECT_LIMIT); |
| 484 | MODULE_DESCRIPTION("nftables limit expression support" ); |
| 485 | |