net/netfilter/nfnetlink_acct.c

   1 /*
   2  * (C) 2011 Pablo Neira Ayuso <pablo@netfilter.org>
   3  * (C) 2011 Intra2net AG <http://www.intra2net.com>
   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 (or any later at your option).
   8  */
   9 #include <linux/init.h>
  10 #include <linux/module.h>
  11 #include <linux/kernel.h>
  12 #include <linux/skbuff.h>
  13 #include <linux/atomic.h>
  14 #include <linux/refcount.h>
  15 #include <linux/netlink.h>
  16 #include <linux/rculist.h>
  17 #include <linux/slab.h>
  18 #include <linux/types.h>
  19 #include <linux/errno.h>
  20 #include <net/netlink.h>
  21 #include <net/sock.h>
  22
  23 #include <linux/netfilter.h>
  24 #include <linux/netfilter/nfnetlink.h>
  25 #include <linux/netfilter/nfnetlink_acct.h>
  26
  27 MODULE_LICENSE("GPL");
  28 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
  29 MODULE_DESCRIPTION("nfacct: Extended Netfilter accounting infrastructure");
  30
  31 struct nf_acct {
  32         atomic64_t              pkts;
  33         atomic64_t              bytes;
  34         unsigned long           flags;
  35         struct list_head        head;
  36         refcount_t              refcnt;
  37         char                    name[NFACCT_NAME_MAX];
  38         struct rcu_head         rcu_head;
  39         char                    data[0];
  40 };
  41
  42 struct nfacct_filter {
  43         u32 value;
  44         u32 mask;
  45 };
  46
  47 #define NFACCT_F_QUOTA (NFACCT_F_QUOTA_PKTS | NFACCT_F_QUOTA_BYTES)
  48 #define NFACCT_OVERQUOTA_BIT    2       /* NFACCT_F_OVERQUOTA */
  49
  50 static int nfnl_acct_new(struct net *net, struct sock *nfnl,
  51                          struct sk_buff *skb, const struct nlmsghdr *nlh,
  52                          const struct nlattr * const tb[])
  53 {
  54         struct nf_acct *nfacct, *matching = NULL;
  55         char *acct_name;
  56         unsigned int size = 0;
  57         u32 flags = 0;
  58
  59         if (!tb[NFACCT_NAME])
  60                 return -EINVAL;
  61
  62         acct_name = nla_data(tb[NFACCT_NAME]);
  63         if (strlen(acct_name) == 0)
  64                 return -EINVAL;
  65
  66         list_for_each_entry(nfacct, &net->nfnl_acct_list, head) {
  67                 if (strncmp(nfacct->name, acct_name, NFACCT_NAME_MAX) != 0)
  68                         continue;
  69
  70                 if (nlh->nlmsg_flags & NLM_F_EXCL)
  71                         return -EEXIST;
  72
  73                 matching = nfacct;
  74                 break;
  75         }
  76
  77         if (matching) {
  78                 if (nlh->nlmsg_flags & NLM_F_REPLACE) {
  79                         /* reset counters if you request a replacement. */
  80                         atomic64_set(&matching->pkts, 0);
  81                         atomic64_set(&matching->bytes, 0);
  82                         smp_mb__before_atomic();
  83                         /* reset overquota flag if quota is enabled. */
  84                         if ((matching->flags & NFACCT_F_QUOTA))
  85                                 clear_bit(NFACCT_OVERQUOTA_BIT,
  86                                           &matching->flags);
  87                         return 0;
  88                 }
  89                 return -EBUSY;
  90         }
  91
  92         if (tb[NFACCT_FLAGS]) {
  93                 flags = ntohl(nla_get_be32(tb[NFACCT_FLAGS]));
  94                 if (flags & ~NFACCT_F_QUOTA)
  95                         return -EOPNOTSUPP;
  96                 if ((flags & NFACCT_F_QUOTA) == NFACCT_F_QUOTA)
  97                         return -EINVAL;
  98                 if (flags & NFACCT_F_OVERQUOTA)
  99                         return -EINVAL;
 100                 if ((flags & NFACCT_F_QUOTA) && !tb[NFACCT_QUOTA])
 101                         return -EINVAL;
 102
 103                 size += sizeof(u64);
 104         }
 105
 106         nfacct = kzalloc(sizeof(struct nf_acct) + size, GFP_KERNEL);
 107         if (nfacct == NULL)
 108                 return -ENOMEM;
 109
 110         if (flags & NFACCT_F_QUOTA) {
 111                 u64 *quota = (u64 *)nfacct->data;
 112
 113                 *quota = be64_to_cpu(nla_get_be64(tb[NFACCT_QUOTA]));
 114                 nfacct->flags = flags;
 115         }
 116
 117         strncpy(nfacct->name, nla_data(tb[NFACCT_NAME]), NFACCT_NAME_MAX);
 118
 119         if (tb[NFACCT_BYTES]) {
 120                 atomic64_set(&nfacct->bytes,
 121                              be64_to_cpu(nla_get_be64(tb[NFACCT_BYTES])));
 122         }
 123         if (tb[NFACCT_PKTS]) {
 124                 atomic64_set(&nfacct->pkts,
 125                              be64_to_cpu(nla_get_be64(tb[NFACCT_PKTS])));
 126         }
 127         refcount_set(&nfacct->refcnt, 1);
 128         list_add_tail_rcu(&nfacct->head, &net->nfnl_acct_list);
 129         return 0;
 130 }
 131
 132 static int
 133 nfnl_acct_fill_info(struct sk_buff *skb, u32 portid, u32 seq, u32 type,
 134                    int event, struct nf_acct *acct)
 135 {
 136         struct nlmsghdr *nlh;
 137         struct nfgenmsg *nfmsg;
 138         unsigned int flags = portid ? NLM_F_MULTI : 0;
 139         u64 pkts, bytes;
 140         u32 old_flags;
 141
 142         event |= NFNL_SUBSYS_ACCT << 8;
 143         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*nfmsg), flags);
 144         if (nlh == NULL)
 145                 goto nlmsg_failure;
 146
 147         nfmsg = nlmsg_data(nlh);
 148         nfmsg->nfgen_family = AF_UNSPEC;
 149         nfmsg->version = NFNETLINK_V0;
 150         nfmsg->res_id = 0;
 151
 152         if (nla_put_string(skb, NFACCT_NAME, acct->name))
 153                 goto nla_put_failure;
 154
 155         old_flags = acct->flags;
 156         if (type == NFNL_MSG_ACCT_GET_CTRZERO) {
 157                 pkts = atomic64_xchg(&acct->pkts, 0);
 158                 bytes = atomic64_xchg(&acct->bytes, 0);
 159                 smp_mb__before_atomic();
 160                 if (acct->flags & NFACCT_F_QUOTA)
 161                         clear_bit(NFACCT_OVERQUOTA_BIT, &acct->flags);
 162         } else {
 163                 pkts = atomic64_read(&acct->pkts);
 164                 bytes = atomic64_read(&acct->bytes);
 165         }
 166         if (nla_put_be64(skb, NFACCT_PKTS, cpu_to_be64(pkts),
 167                          NFACCT_PAD) ||
 168             nla_put_be64(skb, NFACCT_BYTES, cpu_to_be64(bytes),
 169                          NFACCT_PAD) ||
 170             nla_put_be32(skb, NFACCT_USE, htonl(refcount_read(&acct->refcnt))))
 171                 goto nla_put_failure;
 172         if (acct->flags & NFACCT_F_QUOTA) {
 173                 u64 *quota = (u64 *)acct->data;
 174
 175                 if (nla_put_be32(skb, NFACCT_FLAGS, htonl(old_flags)) ||
 176                     nla_put_be64(skb, NFACCT_QUOTA, cpu_to_be64(*quota),
 177                                  NFACCT_PAD))
 178                         goto nla_put_failure;
 179         }
 180         nlmsg_end(skb, nlh);
 181         return skb->len;
 182
 183 nlmsg_failure:
 184 nla_put_failure:
 185         nlmsg_cancel(skb, nlh);
 186         return -1;
 187 }
 188
 189 static int
 190 nfnl_acct_dump(struct sk_buff *skb, struct netlink_callback *cb)
 191 {
 192         struct net *net = sock_net(skb->sk);
 193         struct nf_acct *cur, *last;
 194         const struct nfacct_filter *filter = cb->data;
 195
 196         if (cb->args[2])
 197                 return 0;
 198
 199         last = (struct nf_acct *)cb->args[1];
 200         if (cb->args[1])
 201                 cb->args[1] = 0;
 202
 203         rcu_read_lock();
 204         list_for_each_entry_rcu(cur, &net->nfnl_acct_list, head) {
 205                 if (last) {
 206                         if (cur != last)
 207                                 continue;
 208
 209                         last = NULL;
 210                 }
 211
 212                 if (filter && (cur->flags & filter->mask) != filter->value)
 213                         continue;
 214
 215                 if (nfnl_acct_fill_info(skb, NETLINK_CB(cb->skb).portid,
 216                                        cb->nlh->nlmsg_seq,
 217                                        NFNL_MSG_TYPE(cb->nlh->nlmsg_type),
 218                                        NFNL_MSG_ACCT_NEW, cur) < 0) {
 219                         cb->args[1] = (unsigned long)cur;
 220                         break;
 221                 }
 222         }
 223         if (!cb->args[1])
 224                 cb->args[2] = 1;
 225         rcu_read_unlock();
 226         return skb->len;
 227 }
 228
 229 static int nfnl_acct_done(struct netlink_callback *cb)
 230 {
 231         kfree(cb->data);
 232         return 0;
 233 }
 234
 235 static const struct nla_policy filter_policy[NFACCT_FILTER_MAX + 1] = {
 236         [NFACCT_FILTER_MASK]    = { .type = NLA_U32 },
 237         [NFACCT_FILTER_VALUE]   = { .type = NLA_U32 },
 238 };
 239
 240 static struct nfacct_filter *
 241 nfacct_filter_alloc(const struct nlattr * const attr)
 242 {
 243         struct nfacct_filter *filter;
 244         struct nlattr *tb[NFACCT_FILTER_MAX + 1];
 245         int err;
 246
 247         err = nla_parse_nested(tb, NFACCT_FILTER_MAX, attr, filter_policy,
 248                                NULL);
 249         if (err < 0)
 250                 return ERR_PTR(err);
 251
 252         if (!tb[NFACCT_FILTER_MASK] || !tb[NFACCT_FILTER_VALUE])
 253                 return ERR_PTR(-EINVAL);
 254
 255         filter = kzalloc(sizeof(struct nfacct_filter), GFP_KERNEL);
 256         if (!filter)
 257                 return ERR_PTR(-ENOMEM);
 258
 259         filter->mask = ntohl(nla_get_be32(tb[NFACCT_FILTER_MASK]));
 260         filter->value = ntohl(nla_get_be32(tb[NFACCT_FILTER_VALUE]));
 261
 262         return filter;
 263 }
 264
 265 static int nfnl_acct_get(struct net *net, struct sock *nfnl,
 266                          struct sk_buff *skb, const struct nlmsghdr *nlh,
 267                          const struct nlattr * const tb[])
 268 {
 269         int ret = -ENOENT;
 270         struct nf_acct *cur;
 271         char *acct_name;
 272
 273         if (nlh->nlmsg_flags & NLM_F_DUMP) {
 274                 struct netlink_dump_control c = {
 275                         .dump = nfnl_acct_dump,
 276                         .done = nfnl_acct_done,
 277                 };
 278
 279                 if (tb[NFACCT_FILTER]) {
 280                         struct nfacct_filter *filter;
 281
 282                         filter = nfacct_filter_alloc(tb[NFACCT_FILTER]);
 283                         if (IS_ERR(filter))
 284                                 return PTR_ERR(filter);
 285
 286                         c.data = filter;
 287                 }
 288                 return netlink_dump_start(nfnl, skb, nlh, &c);
 289         }
 290
 291         if (!tb[NFACCT_NAME])
 292                 return -EINVAL;
 293         acct_name = nla_data(tb[NFACCT_NAME]);
 294
 295         list_for_each_entry(cur, &net->nfnl_acct_list, head) {
 296                 struct sk_buff *skb2;
 297
 298                 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
 299                         continue;
 300
 301                 skb2 = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 302                 if (skb2 == NULL) {
 303                         ret = -ENOMEM;
 304                         break;
 305                 }
 306
 307                 ret = nfnl_acct_fill_info(skb2, NETLINK_CB(skb).portid,
 308                                          nlh->nlmsg_seq,
 309                                          NFNL_MSG_TYPE(nlh->nlmsg_type),
 310                                          NFNL_MSG_ACCT_NEW, cur);
 311                 if (ret <= 0) {
 312                         kfree_skb(skb2);
 313                         break;
 314                 }
 315                 ret = netlink_unicast(nfnl, skb2, NETLINK_CB(skb).portid,
 316                                         MSG_DONTWAIT);
 317                 if (ret > 0)
 318                         ret = 0;
 319
 320                 /* this avoids a loop in nfnetlink. */
 321                 return ret == -EAGAIN ? -ENOBUFS : ret;
 322         }
 323         return ret;
 324 }
 325
 326 /* try to delete object, fail if it is still in use. */
 327 static int nfnl_acct_try_del(struct nf_acct *cur)
 328 {
 329         int ret = 0;
 330
 331         /* We want to avoid races with nfnl_acct_put. So only when the current
 332          * refcnt is 1, we decrease it to 0.
 333          */
 334         if (refcount_dec_if_one(&cur->refcnt)) {
 335                 /* We are protected by nfnl mutex. */
 336                 list_del_rcu(&cur->head);
 337                 kfree_rcu(cur, rcu_head);
 338         } else {
 339                 ret = -EBUSY;
 340         }
 341         return ret;
 342 }
 343
 344 static int nfnl_acct_del(struct net *net, struct sock *nfnl,
 345                          struct sk_buff *skb, const struct nlmsghdr *nlh,
 346                          const struct nlattr * const tb[])
 347 {
 348         struct nf_acct *cur, *tmp;
 349         int ret = -ENOENT;
 350         char *acct_name;
 351
 352         if (!tb[NFACCT_NAME]) {
 353                 list_for_each_entry_safe(cur, tmp, &net->nfnl_acct_list, head)
 354                         nfnl_acct_try_del(cur);
 355
 356                 return 0;
 357         }
 358         acct_name = nla_data(tb[NFACCT_NAME]);
 359
 360         list_for_each_entry(cur, &net->nfnl_acct_list, head) {
 361                 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX) != 0)
 362                         continue;
 363
 364                 ret = nfnl_acct_try_del(cur);
 365                 if (ret < 0)
 366                         return ret;
 367
 368                 break;
 369         }
 370         return ret;
 371 }
 372
 373 static const struct nla_policy nfnl_acct_policy[NFACCT_MAX+1] = {
 374         [NFACCT_NAME] = { .type = NLA_NUL_STRING, .len = NFACCT_NAME_MAX-1 },
 375         [NFACCT_BYTES] = { .type = NLA_U64 },
 376         [NFACCT_PKTS] = { .type = NLA_U64 },
 377         [NFACCT_FLAGS] = { .type = NLA_U32 },
 378         [NFACCT_QUOTA] = { .type = NLA_U64 },
 379         [NFACCT_FILTER] = {.type = NLA_NESTED },
 380 };
 381
 382 static const struct nfnl_callback nfnl_acct_cb[NFNL_MSG_ACCT_MAX] = {
 383         [NFNL_MSG_ACCT_NEW]             = { .call = nfnl_acct_new,
 384                                             .attr_count = NFACCT_MAX,
 385                                             .policy = nfnl_acct_policy },
 386         [NFNL_MSG_ACCT_GET]             = { .call = nfnl_acct_get,
 387                                             .attr_count = NFACCT_MAX,
 388                                             .policy = nfnl_acct_policy },
 389         [NFNL_MSG_ACCT_GET_CTRZERO]     = { .call = nfnl_acct_get,
 390                                             .attr_count = NFACCT_MAX,
 391                                             .policy = nfnl_acct_policy },
 392         [NFNL_MSG_ACCT_DEL]             = { .call = nfnl_acct_del,
 393                                             .attr_count = NFACCT_MAX,
 394                                             .policy = nfnl_acct_policy },
 395 };
 396
 397 static const struct nfnetlink_subsystem nfnl_acct_subsys = {
 398         .name                           = "acct",
 399         .subsys_id                      = NFNL_SUBSYS_ACCT,
 400         .cb_count                       = NFNL_MSG_ACCT_MAX,
 401         .cb                             = nfnl_acct_cb,
 402 };
 403
 404 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ACCT);
 405
 406 struct nf_acct *nfnl_acct_find_get(struct net *net, const char *acct_name)
 407 {
 408         struct nf_acct *cur, *acct = NULL;
 409
 410         rcu_read_lock();
 411         list_for_each_entry_rcu(cur, &net->nfnl_acct_list, head) {
 412                 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
 413                         continue;
 414
 415                 if (!try_module_get(THIS_MODULE))
 416                         goto err;
 417
 418                 if (!refcount_inc_not_zero(&cur->refcnt)) {
 419                         module_put(THIS_MODULE);
 420                         goto err;
 421                 }
 422
 423                 acct = cur;
 424                 break;
 425         }
 426 err:
 427         rcu_read_unlock();
 428         return acct;
 429 }
 430 EXPORT_SYMBOL_GPL(nfnl_acct_find_get);
 431
 432 void nfnl_acct_put(struct nf_acct *acct)
 433 {
 434         if (refcount_dec_and_test(&acct->refcnt))
 435                 kfree_rcu(acct, rcu_head);
 436
 437         module_put(THIS_MODULE);
 438 }
 439 EXPORT_SYMBOL_GPL(nfnl_acct_put);
 440
 441 void nfnl_acct_update(const struct sk_buff *skb, struct nf_acct *nfacct)
 442 {
 443         atomic64_inc(&nfacct->pkts);
 444         atomic64_add(skb->len, &nfacct->bytes);
 445 }
 446 EXPORT_SYMBOL_GPL(nfnl_acct_update);
 447
 448 static void nfnl_overquota_report(struct net *net, struct nf_acct *nfacct)
 449 {
 450         int ret;
 451         struct sk_buff *skb;
 452
 453         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
 454         if (skb == NULL)
 455                 return;
 456
 457         ret = nfnl_acct_fill_info(skb, 0, 0, NFNL_MSG_ACCT_OVERQUOTA, 0,
 458                                   nfacct);
 459         if (ret <= 0) {
 460                 kfree_skb(skb);
 461                 return;
 462         }
 463         netlink_broadcast(net->nfnl, skb, 0, NFNLGRP_ACCT_QUOTA,
 464                           GFP_ATOMIC);
 465 }
 466
 467 int nfnl_acct_overquota(struct net *net, const struct sk_buff *skb,
 468                         struct nf_acct *nfacct)
 469 {
 470         u64 now;
 471         u64 *quota;
 472         int ret = NFACCT_UNDERQUOTA;
 473
 474         /* no place here if we don't have a quota */
 475         if (!(nfacct->flags & NFACCT_F_QUOTA))
 476                 return NFACCT_NO_QUOTA;
 477
 478         quota = (u64 *)nfacct->data;
 479         now = (nfacct->flags & NFACCT_F_QUOTA_PKTS) ?
 480                atomic64_read(&nfacct->pkts) : atomic64_read(&nfacct->bytes);
 481
 482         ret = now > *quota;
 483
 484         if (now >= *quota &&
 485             !test_and_set_bit(NFACCT_OVERQUOTA_BIT, &nfacct->flags)) {
 486                 nfnl_overquota_report(net, nfacct);
 487         }
 488
 489         return ret;
 490 }
 491 EXPORT_SYMBOL_GPL(nfnl_acct_overquota);
 492
 493 static int __net_init nfnl_acct_net_init(struct net *net)
 494 {
 495         INIT_LIST_HEAD(&net->nfnl_acct_list);
 496
 497         return 0;
 498 }
 499
 500 static void __net_exit nfnl_acct_net_exit(struct net *net)
 501 {
 502         struct nf_acct *cur, *tmp;
 503
 504         list_for_each_entry_safe(cur, tmp, &net->nfnl_acct_list, head) {
 505                 list_del_rcu(&cur->head);
 506
 507                 if (refcount_dec_and_test(&cur->refcnt))
 508                         kfree_rcu(cur, rcu_head);
 509         }
 510 }
 511
 512 static struct pernet_operations nfnl_acct_ops = {
 513         .init   = nfnl_acct_net_init,
 514         .exit   = nfnl_acct_net_exit,
 515 };
 516
 517 static int __init nfnl_acct_init(void)
 518 {
 519         int ret;
 520
 521         ret = register_pernet_subsys(&nfnl_acct_ops);
 522         if (ret < 0) {
 523                 pr_err("nfnl_acct_init: failed to register pernet ops\n");
 524                 goto err_out;
 525         }
 526
 527         pr_info("nfnl_acct: registering with nfnetlink.\n");
 528         ret = nfnetlink_subsys_register(&nfnl_acct_subsys);
 529         if (ret < 0) {
 530                 pr_err("nfnl_acct_init: cannot register with nfnetlink.\n");
 531                 goto cleanup_pernet;
 532         }
 533         return 0;
 534
 535 cleanup_pernet:
 536         unregister_pernet_subsys(&nfnl_acct_ops);
 537 err_out:
 538         return ret;
 539 }
 540
 541 static void __exit nfnl_acct_exit(void)
 542 {
 543         pr_info("nfnl_acct: unregistering from nfnetlink.\n");
 544         nfnetlink_subsys_unregister(&nfnl_acct_subsys);
 545         unregister_pernet_subsys(&nfnl_acct_ops);
 546 }
 547
 548 module_init(nfnl_acct_init);
 549 module_exit(nfnl_acct_exit);