net/netfilter/nft_ct.c

   1 /*
   2  * Copyright (c) 2008-2009 Patrick McHardy <kaber@trash.net>
   3  * Copyright (c) 2016 Pablo Neira Ayuso <pablo@netfilter.org>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License version 2 as
   7  * published by the Free Software Foundation.
   8  *
   9  * Development of this code funded by Astaro AG (http://www.astaro.com/)
  10  */
  11
  12 #include <linux/kernel.h>
  13 #include <linux/init.h>
  14 #include <linux/module.h>
  15 #include <linux/netlink.h>
  16 #include <linux/netfilter.h>
  17 #include <linux/netfilter/nf_tables.h>
  18 #include <net/netfilter/nf_tables.h>
  19 #include <net/netfilter/nf_conntrack.h>
  20 #include <net/netfilter/nf_conntrack_acct.h>
  21 #include <net/netfilter/nf_conntrack_tuple.h>
  22 #include <net/netfilter/nf_conntrack_helper.h>
  23 #include <net/netfilter/nf_conntrack_ecache.h>
  24 #include <net/netfilter/nf_conntrack_labels.h>
  25
  26 struct nft_ct {
  27         enum nft_ct_keys        key:8;
  28         enum ip_conntrack_dir   dir:8;
  29         union {
  30                 enum nft_registers      dreg:8;
  31                 enum nft_registers      sreg:8;
  32         };
  33 };
  34
  35 static u64 nft_ct_get_eval_counter(const struct nf_conn_counter *c,
  36                                    enum nft_ct_keys k,
  37                                    enum ip_conntrack_dir d)
  38 {
  39         if (d < IP_CT_DIR_MAX)
  40                 return k == NFT_CT_BYTES ? atomic64_read(&c[d].bytes) :
  41                                            atomic64_read(&c[d].packets);
  42
  43         return nft_ct_get_eval_counter(c, k, IP_CT_DIR_ORIGINAL) +
  44                nft_ct_get_eval_counter(c, k, IP_CT_DIR_REPLY);
  45 }
  46
  47 static void nft_ct_get_eval(const struct nft_expr *expr,
  48                             struct nft_regs *regs,
  49                             const struct nft_pktinfo *pkt)
  50 {
  51         const struct nft_ct *priv = nft_expr_priv(expr);
  52         u32 *dest = &regs->data[priv->dreg];
  53         enum ip_conntrack_info ctinfo;
  54         const struct nf_conn *ct;
  55         const struct nf_conn_help *help;
  56         const struct nf_conntrack_tuple *tuple;
  57         const struct nf_conntrack_helper *helper;
  58         unsigned int state;
  59
  60         ct = nf_ct_get(pkt->skb, &ctinfo);
  61
  62         switch (priv->key) {
  63         case NFT_CT_STATE:
  64                 if (ct == NULL)
  65                         state = NF_CT_STATE_INVALID_BIT;
  66                 else if (nf_ct_is_untracked(ct))
  67                         state = NF_CT_STATE_UNTRACKED_BIT;
  68                 else
  69                         state = NF_CT_STATE_BIT(ctinfo);
  70                 *dest = state;
  71                 return;
  72         default:
  73                 break;
  74         }
  75
  76         if (ct == NULL)
  77                 goto err;
  78
  79         switch (priv->key) {
  80         case NFT_CT_DIRECTION:
  81                 *dest = CTINFO2DIR(ctinfo);
  82                 return;
  83         case NFT_CT_STATUS:
  84                 *dest = ct->status;
  85                 return;
  86 #ifdef CONFIG_NF_CONNTRACK_MARK
  87         case NFT_CT_MARK:
  88                 *dest = ct->mark;
  89                 return;
  90 #endif
  91 #ifdef CONFIG_NF_CONNTRACK_SECMARK
  92         case NFT_CT_SECMARK:
  93                 *dest = ct->secmark;
  94                 return;
  95 #endif
  96         case NFT_CT_EXPIRATION:
  97                 *dest = jiffies_to_msecs(nf_ct_expires(ct));
  98                 return;
  99         case NFT_CT_HELPER:
 100                 if (ct->master == NULL)
 101                         goto err;
 102                 help = nfct_help(ct->master);
 103                 if (help == NULL)
 104                         goto err;
 105                 helper = rcu_dereference(help->helper);
 106                 if (helper == NULL)
 107                         goto err;
 108                 strncpy((char *)dest, helper->name, NF_CT_HELPER_NAME_LEN);
 109                 return;
 110 #ifdef CONFIG_NF_CONNTRACK_LABELS
 111         case NFT_CT_LABELS: {
 112                 struct nf_conn_labels *labels = nf_ct_labels_find(ct);
 113
 114                 if (labels)
 115                         memcpy(dest, labels->bits, NF_CT_LABELS_MAX_SIZE);
 116                 else
 117                         memset(dest, 0, NF_CT_LABELS_MAX_SIZE);
 118                 return;
 119         }
 120 #endif
 121         case NFT_CT_BYTES: /* fallthrough */
 122         case NFT_CT_PKTS: {
 123                 const struct nf_conn_acct *acct = nf_conn_acct_find(ct);
 124                 u64 count = 0;
 125
 126                 if (acct)
 127                         count = nft_ct_get_eval_counter(acct->counter,
 128                                                         priv->key, priv->dir);
 129                 memcpy(dest, &count, sizeof(count));
 130                 return;
 131         }
 132         case NFT_CT_L3PROTOCOL:
 133                 *dest = nf_ct_l3num(ct);
 134                 return;
 135         case NFT_CT_PROTOCOL:
 136                 *dest = nf_ct_protonum(ct);
 137                 return;
 138         default:
 139                 break;
 140         }
 141
 142         tuple = &ct->tuplehash[priv->dir].tuple;
 143         switch (priv->key) {
 144         case NFT_CT_SRC:
 145                 memcpy(dest, tuple->src.u3.all,
 146                        nf_ct_l3num(ct) == NFPROTO_IPV4 ? 4 : 16);
 147                 return;
 148         case NFT_CT_DST:
 149                 memcpy(dest, tuple->dst.u3.all,
 150                        nf_ct_l3num(ct) == NFPROTO_IPV4 ? 4 : 16);
 151                 return;
 152         case NFT_CT_PROTO_SRC:
 153                 *dest = (__force __u16)tuple->src.u.all;
 154                 return;
 155         case NFT_CT_PROTO_DST:
 156                 *dest = (__force __u16)tuple->dst.u.all;
 157                 return;
 158         default:
 159                 break;
 160         }
 161         return;
 162 err:
 163         regs->verdict.code = NFT_BREAK;
 164 }
 165
 166 static void nft_ct_set_eval(const struct nft_expr *expr,
 167                             struct nft_regs *regs,
 168                             const struct nft_pktinfo *pkt)
 169 {
 170         const struct nft_ct *priv = nft_expr_priv(expr);
 171         struct sk_buff *skb = pkt->skb;
 172 #ifdef CONFIG_NF_CONNTRACK_MARK
 173         u32 value = regs->data[priv->sreg];
 174 #endif
 175         enum ip_conntrack_info ctinfo;
 176         struct nf_conn *ct;
 177
 178         ct = nf_ct_get(skb, &ctinfo);
 179         if (ct == NULL)
 180                 return;
 181
 182         switch (priv->key) {
 183 #ifdef CONFIG_NF_CONNTRACK_MARK
 184         case NFT_CT_MARK:
 185                 if (ct->mark != value) {
 186                         ct->mark = value;
 187                         nf_conntrack_event_cache(IPCT_MARK, ct);
 188                 }
 189                 break;
 190 #endif
 191 #ifdef CONFIG_NF_CONNTRACK_LABELS
 192         case NFT_CT_LABELS:
 193                 nf_connlabels_replace(ct,
 194                                       &regs->data[priv->sreg],
 195                                       &regs->data[priv->sreg],
 196                                       NF_CT_LABELS_MAX_SIZE / sizeof(u32));
 197                 break;
 198 #endif
 199         default:
 200                 break;
 201         }
 202 }
 203
 204 static const struct nla_policy nft_ct_policy[NFTA_CT_MAX + 1] = {
 205         [NFTA_CT_DREG]          = { .type = NLA_U32 },
 206         [NFTA_CT_KEY]           = { .type = NLA_U32 },
 207         [NFTA_CT_DIRECTION]     = { .type = NLA_U8 },
 208         [NFTA_CT_SREG]          = { .type = NLA_U32 },
 209 };
 210
 211 static int nft_ct_netns_get(struct net *net, uint8_t family)
 212 {
 213         int err;
 214
 215         if (family == NFPROTO_INET) {
 216                 err = nf_ct_netns_get(net, NFPROTO_IPV4);
 217                 if (err < 0)
 218                         goto err1;
 219                 err = nf_ct_netns_get(net, NFPROTO_IPV6);
 220                 if (err < 0)
 221                         goto err2;
 222         } else {
 223                 err = nf_ct_netns_get(net, family);
 224                 if (err < 0)
 225                         goto err1;
 226         }
 227         return 0;
 228
 229 err2:
 230         nf_ct_netns_put(net, NFPROTO_IPV4);
 231 err1:
 232         return err;
 233 }
 234
 235 static void nft_ct_netns_put(struct net *net, uint8_t family)
 236 {
 237         if (family == NFPROTO_INET) {
 238                 nf_ct_netns_put(net, NFPROTO_IPV4);
 239                 nf_ct_netns_put(net, NFPROTO_IPV6);
 240         } else
 241                 nf_ct_netns_put(net, family);
 242 }
 243
 244 static int nft_ct_get_init(const struct nft_ctx *ctx,
 245                            const struct nft_expr *expr,
 246                            const struct nlattr * const tb[])
 247 {
 248         struct nft_ct *priv = nft_expr_priv(expr);
 249         unsigned int len;
 250         int err;
 251
 252         priv->key = ntohl(nla_get_be32(tb[NFTA_CT_KEY]));
 253         switch (priv->key) {
 254         case NFT_CT_DIRECTION:
 255                 if (tb[NFTA_CT_DIRECTION] != NULL)
 256                         return -EINVAL;
 257                 len = sizeof(u8);
 258                 break;
 259         case NFT_CT_STATE:
 260         case NFT_CT_STATUS:
 261 #ifdef CONFIG_NF_CONNTRACK_MARK
 262         case NFT_CT_MARK:
 263 #endif
 264 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 265         case NFT_CT_SECMARK:
 266 #endif
 267         case NFT_CT_EXPIRATION:
 268                 if (tb[NFTA_CT_DIRECTION] != NULL)
 269                         return -EINVAL;
 270                 len = sizeof(u32);
 271                 break;
 272 #ifdef CONFIG_NF_CONNTRACK_LABELS
 273         case NFT_CT_LABELS:
 274                 if (tb[NFTA_CT_DIRECTION] != NULL)
 275                         return -EINVAL;
 276                 len = NF_CT_LABELS_MAX_SIZE;
 277                 break;
 278 #endif
 279         case NFT_CT_HELPER:
 280                 if (tb[NFTA_CT_DIRECTION] != NULL)
 281                         return -EINVAL;
 282                 len = NF_CT_HELPER_NAME_LEN;
 283                 break;
 284
 285         case NFT_CT_L3PROTOCOL:
 286         case NFT_CT_PROTOCOL:
 287                 /* For compatibility, do not report error if NFTA_CT_DIRECTION
 288                  * attribute is specified.
 289                  */
 290                 len = sizeof(u8);
 291                 break;
 292         case NFT_CT_SRC:
 293         case NFT_CT_DST:
 294                 if (tb[NFTA_CT_DIRECTION] == NULL)
 295                         return -EINVAL;
 296
 297                 switch (ctx->afi->family) {
 298                 case NFPROTO_IPV4:
 299                         len = FIELD_SIZEOF(struct nf_conntrack_tuple,
 300                                            src.u3.ip);
 301                         break;
 302                 case NFPROTO_IPV6:
 303                 case NFPROTO_INET:
 304                         len = FIELD_SIZEOF(struct nf_conntrack_tuple,
 305                                            src.u3.ip6);
 306                         break;
 307                 default:
 308                         return -EAFNOSUPPORT;
 309                 }
 310                 break;
 311         case NFT_CT_PROTO_SRC:
 312         case NFT_CT_PROTO_DST:
 313                 if (tb[NFTA_CT_DIRECTION] == NULL)
 314                         return -EINVAL;
 315                 len = FIELD_SIZEOF(struct nf_conntrack_tuple, src.u.all);
 316                 break;
 317         case NFT_CT_BYTES:
 318         case NFT_CT_PKTS:
 319                 /* no direction? return sum of original + reply */
 320                 if (tb[NFTA_CT_DIRECTION] == NULL)
 321                         priv->dir = IP_CT_DIR_MAX;
 322                 len = sizeof(u64);
 323                 break;
 324         default:
 325                 return -EOPNOTSUPP;
 326         }
 327
 328         if (tb[NFTA_CT_DIRECTION] != NULL) {
 329                 priv->dir = nla_get_u8(tb[NFTA_CT_DIRECTION]);
 330                 switch (priv->dir) {
 331                 case IP_CT_DIR_ORIGINAL:
 332                 case IP_CT_DIR_REPLY:
 333                         break;
 334                 default:
 335                         return -EINVAL;
 336                 }
 337         }
 338
 339         priv->dreg = nft_parse_register(tb[NFTA_CT_DREG]);
 340         err = nft_validate_register_store(ctx, priv->dreg, NULL,
 341                                           NFT_DATA_VALUE, len);
 342         if (err < 0)
 343                 return err;
 344
 345         err = nft_ct_netns_get(ctx->net, ctx->afi->family);
 346         if (err < 0)
 347                 return err;
 348
 349         if (priv->key == NFT_CT_BYTES || priv->key == NFT_CT_PKTS)
 350                 nf_ct_set_acct(ctx->net, true);
 351
 352         return 0;
 353 }
 354
 355 static int nft_ct_set_init(const struct nft_ctx *ctx,
 356                            const struct nft_expr *expr,
 357                            const struct nlattr * const tb[])
 358 {
 359         struct nft_ct *priv = nft_expr_priv(expr);
 360         bool label_got = false;
 361         unsigned int len;
 362         int err;
 363
 364         priv->key = ntohl(nla_get_be32(tb[NFTA_CT_KEY]));
 365         switch (priv->key) {
 366 #ifdef CONFIG_NF_CONNTRACK_MARK
 367         case NFT_CT_MARK:
 368                 if (tb[NFTA_CT_DIRECTION])
 369                         return -EINVAL;
 370                 len = FIELD_SIZEOF(struct nf_conn, mark);
 371                 break;
 372 #endif
 373 #ifdef CONFIG_NF_CONNTRACK_LABELS
 374         case NFT_CT_LABELS:
 375                 if (tb[NFTA_CT_DIRECTION])
 376                         return -EINVAL;
 377                 len = NF_CT_LABELS_MAX_SIZE;
 378                 err = nf_connlabels_get(ctx->net, (len * BITS_PER_BYTE) - 1);
 379                 if (err)
 380                         return err;
 381                 label_got = true;
 382                 break;
 383 #endif
 384         default:
 385                 return -EOPNOTSUPP;
 386         }
 387
 388         priv->sreg = nft_parse_register(tb[NFTA_CT_SREG]);
 389         err = nft_validate_register_load(priv->sreg, len);
 390         if (err < 0)
 391                 goto err1;
 392
 393         err = nft_ct_netns_get(ctx->net, ctx->afi->family);
 394         if (err < 0)
 395                 goto err1;
 396
 397         return 0;
 398
 399 err1:
 400         if (label_got)
 401                 nf_connlabels_put(ctx->net);
 402         return err;
 403 }
 404
 405 static void nft_ct_get_destroy(const struct nft_ctx *ctx,
 406                                const struct nft_expr *expr)
 407 {
 408         nf_ct_netns_put(ctx->net, ctx->afi->family);
 409 }
 410
 411 static void nft_ct_set_destroy(const struct nft_ctx *ctx,
 412                                const struct nft_expr *expr)
 413 {
 414         struct nft_ct *priv = nft_expr_priv(expr);
 415
 416         switch (priv->key) {
 417 #ifdef CONFIG_NF_CONNTRACK_LABELS
 418         case NFT_CT_LABELS:
 419                 nf_connlabels_put(ctx->net);
 420                 break;
 421 #endif
 422         default:
 423                 break;
 424         }
 425
 426         nft_ct_netns_put(ctx->net, ctx->afi->family);
 427 }
 428
 429 static int nft_ct_get_dump(struct sk_buff *skb, const struct nft_expr *expr)
 430 {
 431         const struct nft_ct *priv = nft_expr_priv(expr);
 432
 433         if (nft_dump_register(skb, NFTA_CT_DREG, priv->dreg))
 434                 goto nla_put_failure;
 435         if (nla_put_be32(skb, NFTA_CT_KEY, htonl(priv->key)))
 436                 goto nla_put_failure;
 437
 438         switch (priv->key) {
 439         case NFT_CT_SRC:
 440         case NFT_CT_DST:
 441         case NFT_CT_PROTO_SRC:
 442         case NFT_CT_PROTO_DST:
 443                 if (nla_put_u8(skb, NFTA_CT_DIRECTION, priv->dir))
 444                         goto nla_put_failure;
 445                 break;
 446         case NFT_CT_BYTES:
 447         case NFT_CT_PKTS:
 448                 if (priv->dir < IP_CT_DIR_MAX &&
 449                     nla_put_u8(skb, NFTA_CT_DIRECTION, priv->dir))
 450                         goto nla_put_failure;
 451                 break;
 452         default:
 453                 break;
 454         }
 455
 456         return 0;
 457
 458 nla_put_failure:
 459         return -1;
 460 }
 461
 462 static int nft_ct_set_dump(struct sk_buff *skb, const struct nft_expr *expr)
 463 {
 464         const struct nft_ct *priv = nft_expr_priv(expr);
 465
 466         if (nft_dump_register(skb, NFTA_CT_SREG, priv->sreg))
 467                 goto nla_put_failure;
 468         if (nla_put_be32(skb, NFTA_CT_KEY, htonl(priv->key)))
 469                 goto nla_put_failure;
 470         return 0;
 471
 472 nla_put_failure:
 473         return -1;
 474 }
 475
 476 static struct nft_expr_type nft_ct_type;
 477 static const struct nft_expr_ops nft_ct_get_ops = {
 478         .type           = &nft_ct_type,
 479         .size           = NFT_EXPR_SIZE(sizeof(struct nft_ct)),
 480         .eval           = nft_ct_get_eval,
 481         .init           = nft_ct_get_init,
 482         .destroy        = nft_ct_get_destroy,
 483         .dump           = nft_ct_get_dump,
 484 };
 485
 486 static const struct nft_expr_ops nft_ct_set_ops = {
 487         .type           = &nft_ct_type,
 488         .size           = NFT_EXPR_SIZE(sizeof(struct nft_ct)),
 489         .eval           = nft_ct_set_eval,
 490         .init           = nft_ct_set_init,
 491         .destroy        = nft_ct_set_destroy,
 492         .dump           = nft_ct_set_dump,
 493 };
 494
 495 static const struct nft_expr_ops *
 496 nft_ct_select_ops(const struct nft_ctx *ctx,
 497                     const struct nlattr * const tb[])
 498 {
 499         if (tb[NFTA_CT_KEY] == NULL)
 500                 return ERR_PTR(-EINVAL);
 501
 502         if (tb[NFTA_CT_DREG] && tb[NFTA_CT_SREG])
 503                 return ERR_PTR(-EINVAL);
 504
 505         if (tb[NFTA_CT_DREG])
 506                 return &nft_ct_get_ops;
 507
 508         if (tb[NFTA_CT_SREG])
 509                 return &nft_ct_set_ops;
 510
 511         return ERR_PTR(-EINVAL);
 512 }
 513
 514 static struct nft_expr_type nft_ct_type __read_mostly = {
 515         .name           = "ct",
 516         .select_ops     = &nft_ct_select_ops,
 517         .policy         = nft_ct_policy,
 518         .maxattr        = NFTA_CT_MAX,
 519         .owner          = THIS_MODULE,
 520 };
 521
 522 static void nft_notrack_eval(const struct nft_expr *expr,
 523                              struct nft_regs *regs,
 524                              const struct nft_pktinfo *pkt)
 525 {
 526         struct sk_buff *skb = pkt->skb;
 527         enum ip_conntrack_info ctinfo;
 528         struct nf_conn *ct;
 529
 530         ct = nf_ct_get(pkt->skb, &ctinfo);
 531         /* Previously seen (loopback or untracked)?  Ignore. */
 532         if (ct)
 533                 return;
 534
 535         ct = nf_ct_untracked_get();
 536         atomic_inc(&ct->ct_general.use);
 537         skb->nfct = &ct->ct_general;
 538         skb->nfctinfo = IP_CT_NEW;
 539 }
 540
 541 static struct nft_expr_type nft_notrack_type;
 542 static const struct nft_expr_ops nft_notrack_ops = {
 543         .type           = &nft_notrack_type,
 544         .size           = NFT_EXPR_SIZE(0),
 545         .eval           = nft_notrack_eval,
 546 };
 547
 548 static struct nft_expr_type nft_notrack_type __read_mostly = {
 549         .name           = "notrack",
 550         .ops            = &nft_notrack_ops,
 551         .owner          = THIS_MODULE,
 552 };
 553
 554 static int __init nft_ct_module_init(void)
 555 {
 556         int err;
 557
 558         BUILD_BUG_ON(NF_CT_LABELS_MAX_SIZE > NFT_REG_SIZE);
 559
 560         err = nft_register_expr(&nft_ct_type);
 561         if (err < 0)
 562                 return err;
 563
 564         err = nft_register_expr(&nft_notrack_type);
 565         if (err < 0)
 566                 goto err1;
 567
 568         return 0;
 569 err1:
 570         nft_unregister_expr(&nft_ct_type);
 571         return err;
 572 }
 573
 574 static void __exit nft_ct_module_exit(void)
 575 {
 576         nft_unregister_expr(&nft_notrack_type);
 577         nft_unregister_expr(&nft_ct_type);
 578 }
 579
 580 module_init(nft_ct_module_init);
 581 module_exit(nft_ct_module_exit);
 582
 583 MODULE_LICENSE("GPL");
 584 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
 585 MODULE_ALIAS_NFT_EXPR("ct");
 586 MODULE_ALIAS_NFT_EXPR("notrack");