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