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