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