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         if (!tb[NFACCT_FILTER_MASK] || !tb[NFACCT_FILTER_VALUE])
 246                 return ERR_PTR(-EINVAL);
 247
 248         filter = kzalloc(sizeof(struct nfacct_filter), GFP_KERNEL);
 249         if (!filter)
 250                 return ERR_PTR(-ENOMEM);
 251
 252         filter->mask = ntohl(nla_get_be32(tb[NFACCT_FILTER_MASK]));
 253         filter->value = ntohl(nla_get_be32(tb[NFACCT_FILTER_VALUE]));
 254
 255         return filter;
 256 }
 257
 258 static int nfnl_acct_get(struct net *net, struct sock *nfnl,
 259                          struct sk_buff *skb, const struct nlmsghdr *nlh,
 260                          const struct nlattr * const tb[])
 261 {
 262         int ret = -ENOENT;
 263         struct nf_acct *cur;
 264         char *acct_name;
 265
 266         if (nlh->nlmsg_flags & NLM_F_DUMP) {
 267                 struct netlink_dump_control c = {
 268                         .dump = nfnl_acct_dump,
 269                         .done = nfnl_acct_done,
 270                 };
 271
 272                 if (tb[NFACCT_FILTER]) {
 273                         struct nfacct_filter *filter;
 274
 275                         filter = nfacct_filter_alloc(tb[NFACCT_FILTER]);
 276                         if (IS_ERR(filter))
 277                                 return PTR_ERR(filter);
 278
 279                         c.data = filter;
 280                 }
 281                 return netlink_dump_start(nfnl, skb, nlh, &c);
 282         }
 283
 284         if (!tb[NFACCT_NAME])
 285                 return -EINVAL;
 286         acct_name = nla_data(tb[NFACCT_NAME]);
 287
 288         list_for_each_entry(cur, &net->nfnl_acct_list, head) {
 289                 struct sk_buff *skb2;
 290
 291                 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
 292                         continue;
 293
 294                 skb2 = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 295                 if (skb2 == NULL) {
 296                         ret = -ENOMEM;
 297                         break;
 298                 }
 299
 300                 ret = nfnl_acct_fill_info(skb2, NETLINK_CB(skb).portid,
 301                                          nlh->nlmsg_seq,
 302                                          NFNL_MSG_TYPE(nlh->nlmsg_type),
 303                                          NFNL_MSG_ACCT_NEW, cur);
 304                 if (ret <= 0) {
 305                         kfree_skb(skb2);
 306                         break;
 307                 }
 308                 ret = netlink_unicast(nfnl, skb2, NETLINK_CB(skb).portid,
 309                                         MSG_DONTWAIT);
 310                 if (ret > 0)
 311                         ret = 0;
 312
 313                 /* this avoids a loop in nfnetlink. */
 314                 return ret == -EAGAIN ? -ENOBUFS : ret;
 315         }
 316         return ret;
 317 }
 318
 319 /* try to delete object, fail if it is still in use. */
 320 static int nfnl_acct_try_del(struct nf_acct *cur)
 321 {
 322         int ret = 0;
 323
 324         /* we want to avoid races with nfnl_acct_find_get. */
 325         if (atomic_dec_and_test(&cur->refcnt)) {
 326                 /* We are protected by nfnl mutex. */
 327                 list_del_rcu(&cur->head);
 328                 kfree_rcu(cur, rcu_head);
 329         } else {
 330                 /* still in use, restore reference counter. */
 331                 atomic_inc(&cur->refcnt);
 332                 ret = -EBUSY;
 333         }
 334         return ret;
 335 }
 336
 337 static int nfnl_acct_del(struct net *net, struct sock *nfnl,
 338                          struct sk_buff *skb, const struct nlmsghdr *nlh,
 339                          const struct nlattr * const tb[])
 340 {
 341         char *acct_name;
 342         struct nf_acct *cur;
 343         int ret = -ENOENT;
 344
 345         if (!tb[NFACCT_NAME]) {
 346                 list_for_each_entry(cur, &net->nfnl_acct_list, head)
 347                         nfnl_acct_try_del(cur);
 348
 349                 return 0;
 350         }
 351         acct_name = nla_data(tb[NFACCT_NAME]);
 352
 353         list_for_each_entry(cur, &net->nfnl_acct_list, head) {
 354                 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX) != 0)
 355                         continue;
 356
 357                 ret = nfnl_acct_try_del(cur);
 358                 if (ret < 0)
 359                         return ret;
 360
 361                 break;
 362         }
 363         return ret;
 364 }
 365
 366 static const struct nla_policy nfnl_acct_policy[NFACCT_MAX+1] = {
 367         [NFACCT_NAME] = { .type = NLA_NUL_STRING, .len = NFACCT_NAME_MAX-1 },
 368         [NFACCT_BYTES] = { .type = NLA_U64 },
 369         [NFACCT_PKTS] = { .type = NLA_U64 },
 370         [NFACCT_FLAGS] = { .type = NLA_U32 },
 371         [NFACCT_QUOTA] = { .type = NLA_U64 },
 372         [NFACCT_FILTER] = {.type = NLA_NESTED },
 373 };
 374
 375 static const struct nfnl_callback nfnl_acct_cb[NFNL_MSG_ACCT_MAX] = {
 376         [NFNL_MSG_ACCT_NEW]             = { .call = nfnl_acct_new,
 377                                             .attr_count = NFACCT_MAX,
 378                                             .policy = nfnl_acct_policy },
 379         [NFNL_MSG_ACCT_GET]             = { .call = nfnl_acct_get,
 380                                             .attr_count = NFACCT_MAX,
 381                                             .policy = nfnl_acct_policy },
 382         [NFNL_MSG_ACCT_GET_CTRZERO]     = { .call = nfnl_acct_get,
 383                                             .attr_count = NFACCT_MAX,
 384                                             .policy = nfnl_acct_policy },
 385         [NFNL_MSG_ACCT_DEL]             = { .call = nfnl_acct_del,
 386                                             .attr_count = NFACCT_MAX,
 387                                             .policy = nfnl_acct_policy },
 388 };
 389
 390 static const struct nfnetlink_subsystem nfnl_acct_subsys = {
 391         .name                           = "acct",
 392         .subsys_id                      = NFNL_SUBSYS_ACCT,
 393         .cb_count                       = NFNL_MSG_ACCT_MAX,
 394         .cb                             = nfnl_acct_cb,
 395 };
 396
 397 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ACCT);
 398
 399 struct nf_acct *nfnl_acct_find_get(struct net *net, const char *acct_name)
 400 {
 401         struct nf_acct *cur, *acct = NULL;
 402
 403         rcu_read_lock();
 404         list_for_each_entry_rcu(cur, &net->nfnl_acct_list, head) {
 405                 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
 406                         continue;
 407
 408                 if (!try_module_get(THIS_MODULE))
 409                         goto err;
 410
 411                 if (!atomic_inc_not_zero(&cur->refcnt)) {
 412                         module_put(THIS_MODULE);
 413                         goto err;
 414                 }
 415
 416                 acct = cur;
 417                 break;
 418         }
 419 err:
 420         rcu_read_unlock();
 421         return acct;
 422 }
 423 EXPORT_SYMBOL_GPL(nfnl_acct_find_get);
 424
 425 void nfnl_acct_put(struct nf_acct *acct)
 426 {
 427         if (atomic_dec_and_test(&acct->refcnt))
 428                 kfree_rcu(acct, rcu_head);
 429
 430         module_put(THIS_MODULE);
 431 }
 432 EXPORT_SYMBOL_GPL(nfnl_acct_put);
 433
 434 void nfnl_acct_update(const struct sk_buff *skb, struct nf_acct *nfacct)
 435 {
 436         atomic64_inc(&nfacct->pkts);
 437         atomic64_add(skb->len, &nfacct->bytes);
 438 }
 439 EXPORT_SYMBOL_GPL(nfnl_acct_update);
 440
 441 static void nfnl_overquota_report(struct nf_acct *nfacct)
 442 {
 443         int ret;
 444         struct sk_buff *skb;
 445
 446         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
 447         if (skb == NULL)
 448                 return;
 449
 450         ret = nfnl_acct_fill_info(skb, 0, 0, NFNL_MSG_ACCT_OVERQUOTA, 0,
 451                                   nfacct);
 452         if (ret <= 0) {
 453                 kfree_skb(skb);
 454                 return;
 455         }
 456         netlink_broadcast(init_net.nfnl, skb, 0, NFNLGRP_ACCT_QUOTA,
 457                           GFP_ATOMIC);
 458 }
 459
 460 int nfnl_acct_overquota(const struct sk_buff *skb, struct nf_acct *nfacct)
 461 {
 462         u64 now;
 463         u64 *quota;
 464         int ret = NFACCT_UNDERQUOTA;
 465
 466         /* no place here if we don't have a quota */
 467         if (!(nfacct->flags & NFACCT_F_QUOTA))
 468                 return NFACCT_NO_QUOTA;
 469
 470         quota = (u64 *)nfacct->data;
 471         now = (nfacct->flags & NFACCT_F_QUOTA_PKTS) ?
 472                atomic64_read(&nfacct->pkts) : atomic64_read(&nfacct->bytes);
 473
 474         ret = now > *quota;
 475
 476         if (now >= *quota &&
 477             !test_and_set_bit(NFACCT_OVERQUOTA_BIT, &nfacct->flags)) {
 478                 nfnl_overquota_report(nfacct);
 479         }
 480
 481         return ret;
 482 }
 483 EXPORT_SYMBOL_GPL(nfnl_acct_overquota);
 484
 485 static int __net_init nfnl_acct_net_init(struct net *net)
 486 {
 487         INIT_LIST_HEAD(&net->nfnl_acct_list);
 488
 489         return 0;
 490 }
 491
 492 static void __net_exit nfnl_acct_net_exit(struct net *net)
 493 {
 494         struct nf_acct *cur, *tmp;
 495
 496         list_for_each_entry_safe(cur, tmp, &net->nfnl_acct_list, head) {
 497                 list_del_rcu(&cur->head);
 498
 499                 if (atomic_dec_and_test(&cur->refcnt))
 500                         kfree_rcu(cur, rcu_head);
 501         }
 502 }
 503
 504 static struct pernet_operations nfnl_acct_ops = {
 505         .init   = nfnl_acct_net_init,
 506         .exit   = nfnl_acct_net_exit,
 507 };
 508
 509 static int __init nfnl_acct_init(void)
 510 {
 511         int ret;
 512
 513         ret = register_pernet_subsys(&nfnl_acct_ops);
 514         if (ret < 0) {
 515                 pr_err("nfnl_acct_init: failed to register pernet ops\n");
 516                 goto err_out;
 517         }
 518
 519         pr_info("nfnl_acct: registering with nfnetlink.\n");
 520         ret = nfnetlink_subsys_register(&nfnl_acct_subsys);
 521         if (ret < 0) {
 522                 pr_err("nfnl_acct_init: cannot register with nfnetlink.\n");
 523                 goto cleanup_pernet;
 524         }
 525         return 0;
 526
 527 cleanup_pernet:
 528         unregister_pernet_subsys(&nfnl_acct_ops);
 529 err_out:
 530         return ret;
 531 }
 532
 533 static void __exit nfnl_acct_exit(void)
 534 {
 535         pr_info("nfnl_acct: unregistering from nfnetlink.\n");
 536         nfnetlink_subsys_unregister(&nfnl_acct_subsys);
 537         unregister_pernet_subsys(&nfnl_acct_ops);
 538 }
 539
 540 module_init(nfnl_acct_init);
 541 module_exit(nfnl_acct_exit);