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 static LIST_HEAD(nfnl_acct_list);
  31
  32 struct nf_acct {
  33         atomic64_t              pkts;
  34         atomic64_t              bytes;
  35         unsigned long           flags;
  36         struct list_head        head;
  37         atomic_t                refcnt;
  38         char                    name[NFACCT_NAME_MAX];
  39         struct rcu_head         rcu_head;
  40         char                    data[0];
  41 };
  42
  43 #define NFACCT_F_QUOTA (NFACCT_F_QUOTA_PKTS | NFACCT_F_QUOTA_BYTES)
  44
  45 static int
  46 nfnl_acct_new(struct sock *nfnl, struct sk_buff *skb,
  47              const struct nlmsghdr *nlh, const struct nlattr * const tb[])
  48 {
  49         struct nf_acct *nfacct, *matching = NULL;
  50         char *acct_name;
  51         unsigned int size = 0;
  52         u32 flags = 0;
  53
  54         if (!tb[NFACCT_NAME])
  55                 return -EINVAL;
  56
  57         acct_name = nla_data(tb[NFACCT_NAME]);
  58         if (strlen(acct_name) == 0)
  59                 return -EINVAL;
  60
  61         list_for_each_entry(nfacct, &nfnl_acct_list, head) {
  62                 if (strncmp(nfacct->name, acct_name, NFACCT_NAME_MAX) != 0)
  63                         continue;
  64
  65                 if (nlh->nlmsg_flags & NLM_F_EXCL)
  66                         return -EEXIST;
  67
  68                 matching = nfacct;
  69                 break;
  70         }
  71
  72         if (matching) {
  73                 if (nlh->nlmsg_flags & NLM_F_REPLACE) {
  74                         /* reset counters if you request a replacement. */
  75                         atomic64_set(&matching->pkts, 0);
  76                         atomic64_set(&matching->bytes, 0);
  77                         smp_mb__before_atomic();
  78                         /* reset overquota flag if quota is enabled. */
  79                         if ((matching->flags & NFACCT_F_QUOTA))
  80                                 clear_bit(NFACCT_F_OVERQUOTA, &matching->flags);
  81                         return 0;
  82                 }
  83                 return -EBUSY;
  84         }
  85
  86         if (tb[NFACCT_FLAGS]) {
  87                 flags = ntohl(nla_get_be32(tb[NFACCT_FLAGS]));
  88                 if (flags & ~NFACCT_F_QUOTA)
  89                         return -EOPNOTSUPP;
  90                 if ((flags & NFACCT_F_QUOTA) == NFACCT_F_QUOTA)
  91                         return -EINVAL;
  92                 if (flags & NFACCT_F_OVERQUOTA)
  93                         return -EINVAL;
  94
  95                 size += sizeof(u64);
  96         }
  97
  98         nfacct = kzalloc(sizeof(struct nf_acct) + size, GFP_KERNEL);
  99         if (nfacct == NULL)
 100                 return -ENOMEM;
 101
 102         if (flags & NFACCT_F_QUOTA) {
 103                 u64 *quota = (u64 *)nfacct->data;
 104
 105                 *quota = be64_to_cpu(nla_get_be64(tb[NFACCT_QUOTA]));
 106                 nfacct->flags = flags;
 107         }
 108
 109         strncpy(nfacct->name, nla_data(tb[NFACCT_NAME]), NFACCT_NAME_MAX);
 110
 111         if (tb[NFACCT_BYTES]) {
 112                 atomic64_set(&nfacct->bytes,
 113                              be64_to_cpu(nla_get_be64(tb[NFACCT_BYTES])));
 114         }
 115         if (tb[NFACCT_PKTS]) {
 116                 atomic64_set(&nfacct->pkts,
 117                              be64_to_cpu(nla_get_be64(tb[NFACCT_PKTS])));
 118         }
 119         atomic_set(&nfacct->refcnt, 1);
 120         list_add_tail_rcu(&nfacct->head, &nfnl_acct_list);
 121         return 0;
 122 }
 123
 124 static int
 125 nfnl_acct_fill_info(struct sk_buff *skb, u32 portid, u32 seq, u32 type,
 126                    int event, struct nf_acct *acct)
 127 {
 128         struct nlmsghdr *nlh;
 129         struct nfgenmsg *nfmsg;
 130         unsigned int flags = portid ? NLM_F_MULTI : 0;
 131         u64 pkts, bytes;
 132         u32 old_flags;
 133
 134         event |= NFNL_SUBSYS_ACCT << 8;
 135         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*nfmsg), flags);
 136         if (nlh == NULL)
 137                 goto nlmsg_failure;
 138
 139         nfmsg = nlmsg_data(nlh);
 140         nfmsg->nfgen_family = AF_UNSPEC;
 141         nfmsg->version = NFNETLINK_V0;
 142         nfmsg->res_id = 0;
 143
 144         if (nla_put_string(skb, NFACCT_NAME, acct->name))
 145                 goto nla_put_failure;
 146
 147         old_flags = acct->flags;
 148         if (type == NFNL_MSG_ACCT_GET_CTRZERO) {
 149                 pkts = atomic64_xchg(&acct->pkts, 0);
 150                 bytes = atomic64_xchg(&acct->bytes, 0);
 151                 smp_mb__before_atomic();
 152                 if (acct->flags & NFACCT_F_QUOTA)
 153                         clear_bit(NFACCT_F_OVERQUOTA, &acct->flags);
 154         } else {
 155                 pkts = atomic64_read(&acct->pkts);
 156                 bytes = atomic64_read(&acct->bytes);
 157         }
 158         if (nla_put_be64(skb, NFACCT_PKTS, cpu_to_be64(pkts)) ||
 159             nla_put_be64(skb, NFACCT_BYTES, cpu_to_be64(bytes)) ||
 160             nla_put_be32(skb, NFACCT_USE, htonl(atomic_read(&acct->refcnt))))
 161                 goto nla_put_failure;
 162         if (acct->flags & NFACCT_F_QUOTA) {
 163                 u64 *quota = (u64 *)acct->data;
 164
 165                 if (nla_put_be32(skb, NFACCT_FLAGS, htonl(old_flags)) ||
 166                     nla_put_be64(skb, NFACCT_QUOTA, cpu_to_be64(*quota)))
 167                         goto nla_put_failure;
 168         }
 169         nlmsg_end(skb, nlh);
 170         return skb->len;
 171
 172 nlmsg_failure:
 173 nla_put_failure:
 174         nlmsg_cancel(skb, nlh);
 175         return -1;
 176 }
 177
 178 static int
 179 nfnl_acct_dump(struct sk_buff *skb, struct netlink_callback *cb)
 180 {
 181         struct nf_acct *cur, *last;
 182
 183         if (cb->args[2])
 184                 return 0;
 185
 186         last = (struct nf_acct *)cb->args[1];
 187         if (cb->args[1])
 188                 cb->args[1] = 0;
 189
 190         rcu_read_lock();
 191         list_for_each_entry_rcu(cur, &nfnl_acct_list, head) {
 192                 if (last) {
 193                         if (cur != last)
 194                                 continue;
 195
 196                         last = NULL;
 197                 }
 198                 if (nfnl_acct_fill_info(skb, NETLINK_CB(cb->skb).portid,
 199                                        cb->nlh->nlmsg_seq,
 200                                        NFNL_MSG_TYPE(cb->nlh->nlmsg_type),
 201                                        NFNL_MSG_ACCT_NEW, cur) < 0) {
 202                         cb->args[1] = (unsigned long)cur;
 203                         break;
 204                 }
 205         }
 206         if (!cb->args[1])
 207                 cb->args[2] = 1;
 208         rcu_read_unlock();
 209         return skb->len;
 210 }
 211
 212 static int
 213 nfnl_acct_get(struct sock *nfnl, struct sk_buff *skb,
 214              const struct nlmsghdr *nlh, const struct nlattr * const tb[])
 215 {
 216         int ret = -ENOENT;
 217         struct nf_acct *cur;
 218         char *acct_name;
 219
 220         if (nlh->nlmsg_flags & NLM_F_DUMP) {
 221                 struct netlink_dump_control c = {
 222                         .dump = nfnl_acct_dump,
 223                 };
 224                 return netlink_dump_start(nfnl, skb, nlh, &c);
 225         }
 226
 227         if (!tb[NFACCT_NAME])
 228                 return -EINVAL;
 229         acct_name = nla_data(tb[NFACCT_NAME]);
 230
 231         list_for_each_entry(cur, &nfnl_acct_list, head) {
 232                 struct sk_buff *skb2;
 233
 234                 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
 235                         continue;
 236
 237                 skb2 = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 238                 if (skb2 == NULL) {
 239                         ret = -ENOMEM;
 240                         break;
 241                 }
 242
 243                 ret = nfnl_acct_fill_info(skb2, NETLINK_CB(skb).portid,
 244                                          nlh->nlmsg_seq,
 245                                          NFNL_MSG_TYPE(nlh->nlmsg_type),
 246                                          NFNL_MSG_ACCT_NEW, cur);
 247                 if (ret <= 0) {
 248                         kfree_skb(skb2);
 249                         break;
 250                 }
 251                 ret = netlink_unicast(nfnl, skb2, NETLINK_CB(skb).portid,
 252                                         MSG_DONTWAIT);
 253                 if (ret > 0)
 254                         ret = 0;
 255
 256                 /* this avoids a loop in nfnetlink. */
 257                 return ret == -EAGAIN ? -ENOBUFS : ret;
 258         }
 259         return ret;
 260 }
 261
 262 /* try to delete object, fail if it is still in use. */
 263 static int nfnl_acct_try_del(struct nf_acct *cur)
 264 {
 265         int ret = 0;
 266
 267         /* we want to avoid races with nfnl_acct_find_get. */
 268         if (atomic_dec_and_test(&cur->refcnt)) {
 269                 /* We are protected by nfnl mutex. */
 270                 list_del_rcu(&cur->head);
 271                 kfree_rcu(cur, rcu_head);
 272         } else {
 273                 /* still in use, restore reference counter. */
 274                 atomic_inc(&cur->refcnt);
 275                 ret = -EBUSY;
 276         }
 277         return ret;
 278 }
 279
 280 static int
 281 nfnl_acct_del(struct sock *nfnl, struct sk_buff *skb,
 282              const struct nlmsghdr *nlh, const struct nlattr * const tb[])
 283 {
 284         char *acct_name;
 285         struct nf_acct *cur;
 286         int ret = -ENOENT;
 287
 288         if (!tb[NFACCT_NAME]) {
 289                 list_for_each_entry(cur, &nfnl_acct_list, head)
 290                         nfnl_acct_try_del(cur);
 291
 292                 return 0;
 293         }
 294         acct_name = nla_data(tb[NFACCT_NAME]);
 295
 296         list_for_each_entry(cur, &nfnl_acct_list, head) {
 297                 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX) != 0)
 298                         continue;
 299
 300                 ret = nfnl_acct_try_del(cur);
 301                 if (ret < 0)
 302                         return ret;
 303
 304                 break;
 305         }
 306         return ret;
 307 }
 308
 309 static const struct nla_policy nfnl_acct_policy[NFACCT_MAX+1] = {
 310         [NFACCT_NAME] = { .type = NLA_NUL_STRING, .len = NFACCT_NAME_MAX-1 },
 311         [NFACCT_BYTES] = { .type = NLA_U64 },
 312         [NFACCT_PKTS] = { .type = NLA_U64 },
 313         [NFACCT_FLAGS] = { .type = NLA_U32 },
 314         [NFACCT_QUOTA] = { .type = NLA_U64 },
 315 };
 316
 317 static const struct nfnl_callback nfnl_acct_cb[NFNL_MSG_ACCT_MAX] = {
 318         [NFNL_MSG_ACCT_NEW]             = { .call = nfnl_acct_new,
 319                                             .attr_count = NFACCT_MAX,
 320                                             .policy = nfnl_acct_policy },
 321         [NFNL_MSG_ACCT_GET]             = { .call = nfnl_acct_get,
 322                                             .attr_count = NFACCT_MAX,
 323                                             .policy = nfnl_acct_policy },
 324         [NFNL_MSG_ACCT_GET_CTRZERO]     = { .call = nfnl_acct_get,
 325                                             .attr_count = NFACCT_MAX,
 326                                             .policy = nfnl_acct_policy },
 327         [NFNL_MSG_ACCT_DEL]             = { .call = nfnl_acct_del,
 328                                             .attr_count = NFACCT_MAX,
 329                                             .policy = nfnl_acct_policy },
 330 };
 331
 332 static const struct nfnetlink_subsystem nfnl_acct_subsys = {
 333         .name                           = "acct",
 334         .subsys_id                      = NFNL_SUBSYS_ACCT,
 335         .cb_count                       = NFNL_MSG_ACCT_MAX,
 336         .cb                             = nfnl_acct_cb,
 337 };
 338
 339 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ACCT);
 340
 341 struct nf_acct *nfnl_acct_find_get(const char *acct_name)
 342 {
 343         struct nf_acct *cur, *acct = NULL;
 344
 345         rcu_read_lock();
 346         list_for_each_entry_rcu(cur, &nfnl_acct_list, head) {
 347                 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
 348                         continue;
 349
 350                 if (!try_module_get(THIS_MODULE))
 351                         goto err;
 352
 353                 if (!atomic_inc_not_zero(&cur->refcnt)) {
 354                         module_put(THIS_MODULE);
 355                         goto err;
 356                 }
 357
 358                 acct = cur;
 359                 break;
 360         }
 361 err:
 362         rcu_read_unlock();
 363         return acct;
 364 }
 365 EXPORT_SYMBOL_GPL(nfnl_acct_find_get);
 366
 367 void nfnl_acct_put(struct nf_acct *acct)
 368 {
 369         atomic_dec(&acct->refcnt);
 370         module_put(THIS_MODULE);
 371 }
 372 EXPORT_SYMBOL_GPL(nfnl_acct_put);
 373
 374 void nfnl_acct_update(const struct sk_buff *skb, struct nf_acct *nfacct)
 375 {
 376         atomic64_inc(&nfacct->pkts);
 377         atomic64_add(skb->len, &nfacct->bytes);
 378 }
 379 EXPORT_SYMBOL_GPL(nfnl_acct_update);
 380
 381 static void nfnl_overquota_report(struct nf_acct *nfacct)
 382 {
 383         int ret;
 384         struct sk_buff *skb;
 385
 386         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
 387         if (skb == NULL)
 388                 return;
 389
 390         ret = nfnl_acct_fill_info(skb, 0, 0, NFNL_MSG_ACCT_OVERQUOTA, 0,
 391                                   nfacct);
 392         if (ret <= 0) {
 393                 kfree_skb(skb);
 394                 return;
 395         }
 396         netlink_broadcast(init_net.nfnl, skb, 0, NFNLGRP_ACCT_QUOTA,
 397                           GFP_ATOMIC);
 398 }
 399
 400 int nfnl_acct_overquota(const struct sk_buff *skb, struct nf_acct *nfacct)
 401 {
 402         u64 now;
 403         u64 *quota;
 404         int ret = NFACCT_UNDERQUOTA;
 405
 406         /* no place here if we don't have a quota */
 407         if (!(nfacct->flags & NFACCT_F_QUOTA))
 408                 return NFACCT_NO_QUOTA;
 409
 410         quota = (u64 *)nfacct->data;
 411         now = (nfacct->flags & NFACCT_F_QUOTA_PKTS) ?
 412                atomic64_read(&nfacct->pkts) : atomic64_read(&nfacct->bytes);
 413
 414         ret = now > *quota;
 415
 416         if (now >= *quota &&
 417             !test_and_set_bit(NFACCT_F_OVERQUOTA, &nfacct->flags)) {
 418                 nfnl_overquota_report(nfacct);
 419         }
 420
 421         return ret;
 422 }
 423 EXPORT_SYMBOL_GPL(nfnl_acct_overquota);
 424
 425 static int __init nfnl_acct_init(void)
 426 {
 427         int ret;
 428
 429         pr_info("nfnl_acct: registering with nfnetlink.\n");
 430         ret = nfnetlink_subsys_register(&nfnl_acct_subsys);
 431         if (ret < 0) {
 432                 pr_err("nfnl_acct_init: cannot register with nfnetlink.\n");
 433                 goto err_out;
 434         }
 435         return 0;
 436 err_out:
 437         return ret;
 438 }
 439
 440 static void __exit nfnl_acct_exit(void)
 441 {
 442         struct nf_acct *cur, *tmp;
 443
 444         pr_info("nfnl_acct: unregistering from nfnetlink.\n");
 445         nfnetlink_subsys_unregister(&nfnl_acct_subsys);
 446
 447         list_for_each_entry_safe(cur, tmp, &nfnl_acct_list, head) {
 448                 list_del_rcu(&cur->head);
 449                 /* We are sure that our objects have no clients at this point,
 450                  * it's safe to release them all without checking refcnt. */
 451                 kfree_rcu(cur, rcu_head);
 452         }
 453 }
 454
 455 module_init(nfnl_acct_init);
 456 module_exit(nfnl_acct_exit);