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