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
  50 nfnl_acct_new(struct sock *nfnl, struct sk_buff *skb,
  51              const struct nlmsghdr *nlh, const struct nlattr * const tb[])
  52 {
  53         struct nf_acct *nfacct, *matching = NULL;
  54         struct net *net = sock_net(nfnl);
  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         atomic_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             nla_put_be64(skb, NFACCT_BYTES, cpu_to_be64(bytes)) ||
 168             nla_put_be32(skb, NFACCT_USE, htonl(atomic_read(&acct->refcnt))))
 169                 goto nla_put_failure;
 170         if (acct->flags & NFACCT_F_QUOTA) {
 171                 u64 *quota = (u64 *)acct->data;
 172
 173                 if (nla_put_be32(skb, NFACCT_FLAGS, htonl(old_flags)) ||
 174                     nla_put_be64(skb, NFACCT_QUOTA, cpu_to_be64(*quota)))
 175                         goto nla_put_failure;
 176         }
 177         nlmsg_end(skb, nlh);
 178         return skb->len;
 179
 180 nlmsg_failure:
 181 nla_put_failure:
 182         nlmsg_cancel(skb, nlh);
 183         return -1;
 184 }
 185
 186 static int
 187 nfnl_acct_dump(struct sk_buff *skb, struct netlink_callback *cb)
 188 {
 189         struct net *net = sock_net(skb->sk);
 190         struct nf_acct *cur, *last;
 191         const struct nfacct_filter *filter = cb->data;
 192
 193         if (cb->args[2])
 194                 return 0;
 195
 196         last = (struct nf_acct *)cb->args[1];
 197         if (cb->args[1])
 198                 cb->args[1] = 0;
 199
 200         rcu_read_lock();
 201         list_for_each_entry_rcu(cur, &net->nfnl_acct_list, head) {
 202                 if (last) {
 203                         if (cur != last)
 204                                 continue;
 205
 206                         last = NULL;
 207                 }
 208
 209                 if (filter && (cur->flags & filter->mask) != filter->value)
 210                         continue;
 211
 212                 if (nfnl_acct_fill_info(skb, NETLINK_CB(cb->skb).portid,
 213                                        cb->nlh->nlmsg_seq,
 214                                        NFNL_MSG_TYPE(cb->nlh->nlmsg_type),
 215                                        NFNL_MSG_ACCT_NEW, cur) < 0) {
 216                         cb->args[1] = (unsigned long)cur;
 217                         break;
 218                 }
 219         }
 220         if (!cb->args[1])
 221                 cb->args[2] = 1;
 222         rcu_read_unlock();
 223         return skb->len;
 224 }
 225
 226 static int nfnl_acct_done(struct netlink_callback *cb)
 227 {
 228         kfree(cb->data);
 229         return 0;
 230 }
 231
 232 static const struct nla_policy filter_policy[NFACCT_FILTER_MAX + 1] = {
 233         [NFACCT_FILTER_MASK]    = { .type = NLA_U32 },
 234         [NFACCT_FILTER_VALUE]   = { .type = NLA_U32 },
 235 };
 236
 237 static struct nfacct_filter *
 238 nfacct_filter_alloc(const struct nlattr * const attr)
 239 {
 240         struct nfacct_filter *filter;
 241         struct nlattr *tb[NFACCT_FILTER_MAX + 1];
 242         int err;
 243
 244         err = nla_parse_nested(tb, NFACCT_FILTER_MAX, attr, filter_policy);
 245         if (err < 0)
 246                 return ERR_PTR(err);
 247
 248         if (!tb[NFACCT_FILTER_MASK] || !tb[NFACCT_FILTER_VALUE])
 249                 return ERR_PTR(-EINVAL);
 250
 251         filter = kzalloc(sizeof(struct nfacct_filter), GFP_KERNEL);
 252         if (!filter)
 253                 return ERR_PTR(-ENOMEM);
 254
 255         filter->mask = ntohl(nla_get_be32(tb[NFACCT_FILTER_MASK]));
 256         filter->value = ntohl(nla_get_be32(tb[NFACCT_FILTER_VALUE]));
 257
 258         return filter;
 259 }
 260
 261 static int
 262 nfnl_acct_get(struct sock *nfnl, struct sk_buff *skb,
 263              const struct nlmsghdr *nlh, const struct nlattr * const tb[])
 264 {
 265         struct net *net = sock_net(nfnl);
 266         int ret = -ENOENT;
 267         struct nf_acct *cur;
 268         char *acct_name;
 269
 270         if (nlh->nlmsg_flags & NLM_F_DUMP) {
 271                 struct netlink_dump_control c = {
 272                         .dump = nfnl_acct_dump,
 273                         .done = nfnl_acct_done,
 274                 };
 275
 276                 if (tb[NFACCT_FILTER]) {
 277                         struct nfacct_filter *filter;
 278
 279                         filter = nfacct_filter_alloc(tb[NFACCT_FILTER]);
 280                         if (IS_ERR(filter))
 281                                 return PTR_ERR(filter);
 282
 283                         c.data = filter;
 284                 }
 285                 return netlink_dump_start(nfnl, skb, nlh, &c);
 286         }
 287
 288         if (!tb[NFACCT_NAME])
 289                 return -EINVAL;
 290         acct_name = nla_data(tb[NFACCT_NAME]);
 291
 292         list_for_each_entry(cur, &net->nfnl_acct_list, head) {
 293                 struct sk_buff *skb2;
 294
 295                 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
 296                         continue;
 297
 298                 skb2 = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 299                 if (skb2 == NULL) {
 300                         ret = -ENOMEM;
 301                         break;
 302                 }
 303
 304                 ret = nfnl_acct_fill_info(skb2, NETLINK_CB(skb).portid,
 305                                          nlh->nlmsg_seq,
 306                                          NFNL_MSG_TYPE(nlh->nlmsg_type),
 307                                          NFNL_MSG_ACCT_NEW, cur);
 308                 if (ret <= 0) {
 309                         kfree_skb(skb2);
 310                         break;
 311                 }
 312                 ret = netlink_unicast(nfnl, skb2, NETLINK_CB(skb).portid,
 313                                         MSG_DONTWAIT);
 314                 if (ret > 0)
 315                         ret = 0;
 316
 317                 /* this avoids a loop in nfnetlink. */
 318                 return ret == -EAGAIN ? -ENOBUFS : ret;
 319         }
 320         return ret;
 321 }
 322
 323 /* try to delete object, fail if it is still in use. */
 324 static int nfnl_acct_try_del(struct nf_acct *cur)
 325 {
 326         int ret = 0;
 327
 328         /* we want to avoid races with nfnl_acct_find_get. */
 329         if (atomic_dec_and_test(&cur->refcnt)) {
 330                 /* We are protected by nfnl mutex. */
 331                 list_del_rcu(&cur->head);
 332                 kfree_rcu(cur, rcu_head);
 333         } else {
 334                 /* still in use, restore reference counter. */
 335                 atomic_inc(&cur->refcnt);
 336                 ret = -EBUSY;
 337         }
 338         return ret;
 339 }
 340
 341 static int
 342 nfnl_acct_del(struct sock *nfnl, struct sk_buff *skb,
 343              const struct nlmsghdr *nlh, const struct nlattr * const tb[])
 344 {
 345         struct net *net = sock_net(nfnl);
 346         char *acct_name;
 347         struct nf_acct *cur;
 348         int ret = -ENOENT;
 349
 350         if (!tb[NFACCT_NAME]) {
 351                 list_for_each_entry(cur, &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 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(init_net.nfnl, skb, 0, NFNLGRP_ACCT_QUOTA,
 462                           GFP_ATOMIC);
 463 }
 464
 465 int nfnl_acct_overquota(const struct sk_buff *skb, struct nf_acct *nfacct)
 466 {
 467         u64 now;
 468         u64 *quota;
 469         int ret = NFACCT_UNDERQUOTA;
 470
 471         /* no place here if we don't have a quota */
 472         if (!(nfacct->flags & NFACCT_F_QUOTA))
 473                 return NFACCT_NO_QUOTA;
 474
 475         quota = (u64 *)nfacct->data;
 476         now = (nfacct->flags & NFACCT_F_QUOTA_PKTS) ?
 477                atomic64_read(&nfacct->pkts) : atomic64_read(&nfacct->bytes);
 478
 479         ret = now > *quota;
 480
 481         if (now >= *quota &&
 482             !test_and_set_bit(NFACCT_OVERQUOTA_BIT, &nfacct->flags)) {
 483                 nfnl_overquota_report(nfacct);
 484         }
 485
 486         return ret;
 487 }
 488 EXPORT_SYMBOL_GPL(nfnl_acct_overquota);
 489
 490 static int __net_init nfnl_acct_net_init(struct net *net)
 491 {
 492         INIT_LIST_HEAD(&net->nfnl_acct_list);
 493
 494         return 0;
 495 }
 496
 497 static void __net_exit nfnl_acct_net_exit(struct net *net)
 498 {
 499         struct nf_acct *cur, *tmp;
 500
 501         list_for_each_entry_safe(cur, tmp, &net->nfnl_acct_list, head) {
 502                 list_del_rcu(&cur->head);
 503
 504                 if (atomic_dec_and_test(&cur->refcnt))
 505                         kfree_rcu(cur, rcu_head);
 506         }
 507 }
 508
 509 static struct pernet_operations nfnl_acct_ops = {
 510         .init   = nfnl_acct_net_init,
 511         .exit   = nfnl_acct_net_exit,
 512 };
 513
 514 static int __init nfnl_acct_init(void)
 515 {
 516         int ret;
 517
 518         ret = register_pernet_subsys(&nfnl_acct_ops);
 519         if (ret < 0) {
 520                 pr_err("nfnl_acct_init: failed to register pernet ops\n");
 521                 goto err_out;
 522         }
 523
 524         pr_info("nfnl_acct: registering with nfnetlink.\n");
 525         ret = nfnetlink_subsys_register(&nfnl_acct_subsys);
 526         if (ret < 0) {
 527                 pr_err("nfnl_acct_init: cannot register with nfnetlink.\n");
 528                 goto cleanup_pernet;
 529         }
 530         return 0;
 531
 532 cleanup_pernet:
 533         unregister_pernet_subsys(&nfnl_acct_ops);
 534 err_out:
 535         return ret;
 536 }
 537
 538 static void __exit nfnl_acct_exit(void)
 539 {
 540         pr_info("nfnl_acct: unregistering from nfnetlink.\n");
 541         nfnetlink_subsys_unregister(&nfnl_acct_subsys);
 542         unregister_pernet_subsys(&nfnl_acct_ops);
 543 }
 544
 545 module_init(nfnl_acct_init);
 546 module_exit(nfnl_acct_exit);